Federal Reserve Bank of New York
December 2013 Economic Volume 19 Number 2 19 Number Volume Policy Review
1 Shadow Banking Zoltan Pozsar, Tobias Adrian, Adam Ashcraft, and Hayley Boesky
17 The Rising Gap between Primary and Secondary Mortgage Rates Andreas Fuster, Laurie Goodman, David Lucca, Laurel Madar, Linsey Molloy, and Paul Willen
41 Precarious Slopes? The Great Recession, Federal Stimulus, and New Jersey Schools Rajashri Chakrabarti and Sarah Sutherland
67 The Financial Market Effect of FOMC Minutes Carlo Rosa Economic Policy Review
Editor Kenneth D. Garbade
Coeditors Meta Brown Richard Crump Marco Del Negro Jan Groen Andrew Haughwout Stavros Peristiani
Editorial Staff Valerie LaPorte Michelle Bailer Karen Carter Mike De Mott Anna Snider
Production Staff Jessica Iannuzzi David Rosenberg Jane Urry
Federal Reserve Bank of New York research publications—the Economic Policy Review, Current Issues in Economics and Finance, Staff Reports, Research Update, and the annual publications catalogue—are available at no charge from the Research Group’s website. www.newyorkfed.org/research Follow us on Twitter: @NYFedResearch Federal Reserve Bank of New York Economic Policy Review
December 2013 Volume 19 Number 2
Contents Articles: 1 Shadow Banking Zoltan Pozsar, Tobias Adrian, Adam Ashcraft, and Hayley Boesky
The rapid growth of the market-based financial system since the mid-1980s has changed the nature of financial intermediation. Within the system, “shadow banks” have served a critical role, especially in the run-up to the recent financial crisis. Shadow banks are financial intermediaries that conduct maturity, credit, and liquidity transformation without explicit access to central bank liquidity or public sector credit guarantees. This article documents the institutional features of shadow banks, discusses the banks’ economic roles, and analyzes their relation to the traditional banking system. The authors argue that an understanding of the “plumbing” of the shadow banking system is an important underpinning for any study of financial system interlinkages. They observe that while many current and future reform efforts are focused on remediating the excesses of the recent credit bubble, increased capital and liquidity standards for depository institutions and insurance companies are likely to heighten the returns to shadow banking activity. Thus, shadow banking is expected to be a significant part of the financial system, although very likely in a different form, for the foreseeable future.
17 The Rising Gap between Primary and Secondary Mortgage Rates Andreas Fuster, Laurie Goodman, David Lucca, Laurel Madar, Linsey Molloy, and Paul Willen
While mortgage rates reached historic lows during 2012, the spread between primary and secondary rates rose to very high levels. This trend reflected a number of factors that potentially affected mortgage originator costs and profits and restrained the pass-through from lower secondary rates to borrowers’ funding costs. This article describes the mortgage origination and securitization process and the way in which originator profits are determined. The authors calculate a series of originator profits and unmeasured costs (OPUCs) for the period 1994 to 2012, and show that these OPUCs increased significantly between 2008 and 2012. They also evaluate the extent to which some commonly cited factors, such as changes in loan putback risk, mortgage servicing rights values, and pipeline hedging costs, contributed to the rise in OPUCs. Although some costs of mortgage origination may have risen in recent years, a large component of the rise in OPUCs remains unexplained, pointing to increased profitability of originations. The authors conclude by discussing possible drivers of the rise in profitability, such as capacity constraints and originators’ pricing power resulting from borrowers’ switching costs.
41 Precarious Slopes? The Great Recession, Federal Stimulus, and New Jersey Schools Rajashri Chakrabarti and Sarah Sutherland
While only a sparse literature investigates the impact of the Great Recession on various sectors of the economy, there is virtually no research on the effect on schools. This article starts to fill the void. The authors make use of rich panel data and a trend-shift analysis to study how New Jersey school finances were affected by the onset of the recession and the federal stimulus that followed. Their results show strong evidence of downward shifts in total school funding and expenditures, relative to trend, following the recession. Support of more than $2 billion in American Recovery and Reinvestment Act funding seems to provide a cushion in 2010: While funding and expenditures still fall relative to pre-recession levels, they decline less than in 2009. The infusion of federal funding coincides with significant cuts in state and local support, and the authors mark sharp changes in New Jersey’s relative reliance on the three sources of aid. An examination of the compositional shift in expenditures suggests that the stimulus may have prevented declines in categories linked most closely to instruction. Still, budgetary stress seems to have led to sizable layoffs of nontenured teachers, resulting in an increase in median teacher salary and median experience level. Furthermore, high-poverty and urban school districts were found to sustain larger resource declines than more affluent and less populated districts did in the post-recession era. The study’s findings offer valuable insight into school finances during recessions and can serve as a guide to aid future policy decisions.
67 The Financial Market Effect of Central Bank Minutes Carlo Rosa
The influence of the Federal Reserve’s unanticipated target rate decisions on U.S. asset prices has been the subject of numerous studies. More recently, researchers have looked at the asset price response to statements issued by the Federal Open Market Committee (FOMC). Yet, despite a vast and growing body of evidence on the financial market effect of monetary news released on FOMC meeting days, little is known about the real-time response of U.S. asset prices to the information contained in central bank minutes. This article fills the gap by using a novel, high- frequency data set to examine the effect of the FOMC minutes release on U.S. asset prices— Treasury rates, stock prices, and U.S. dollar exchange rates. The author shows that the release significantly affects the volatility of U.S. asset prices and their trading volume. The magnitude of the effects is economically and statistically significant, and it is similar in magnitude to the Institute for Supply Management manufacturing index release, although smaller than that of the FOMC statement and nonfarm payrolls releases. The asset price response to the minutes, however, has declined in recent years, suggesting that the FOMC has become more transparent by releasing information in a timelier manner.
Zoltan Pozsar, Tobias Adrian, Adam Ashcraft, and Hayley Boesky
Shadow Banking
• Shadow banks are financial intermediaries 1.Introduction that conduct maturity, credit, and liquidity transformation without explicit access to central hadow banking activities consist of credit, maturity, bank liquidity or public sector credit guarantees. Sand liquidity transformation that take place without direct and explicit access to public sources of liquidity or • The banks have played a key role in the credit backstops. These activities are conducted by specialized market-based financial system, particularly financial intermediaries called shadow banks, which are in the run-up to the financial crisis. bound together along an intermediation chain known as the shadow banking system (see “The Shadow Banking 1 • This study describes the institutional features System” Online Appendix). of shadow banks, their economic roles, and In the shadow banking system, credit is intermediated through a wide range of securitization and secured funding their relation to traditional banks. techniques, including asset-backed commercial paper (CP), asset-backed securities (ABS), collateralized debt obligations • The authors suggest that increased capital and (CDOs), and repurchase agreements (repos). While we believe liquidity standards for depository institutions the term “shadow banking,” coined by McCulley (2007), to be and insurance companies will likely heighten a somewhat pejorative name for such a large and important the returns to shadow banking activity. part of the financial system, we have adopted it for use here. Prior to the 2007-09 financial crisis, the shadow banking • Shadow banking, in some form or another, is system provided credit by issuing liquid, short-term liabilities therefore expected to be an important part of against risky, long-term, and often opaque assets. The large the financial system for the foreseeable future. amounts of credit intermediation provided by the shadow banking system contributed to asset price appreciation in residential and commercial real estate markets prior to the financial crisis and to the expansion of credit more generally.
1 This article is complemented by a series of online appendixes (listed in the box on the next page).
Zoltan Pozsar is a senior advisor at the U.S. Treasury Department’s Office This study was first published as Federal Reserve Bank of New York staff of Financial Research; Tobias Adrian is a vice president and Adam Ashcraft report no. 458, July 2010 (www.newyorkfed.org/research/staff_reports/ a senior vice president at the Federal Reserve Bank of New York; sr458.html). The views expressed are those of the authors and do not Hayley Boesky is a vice chairman at Bank of America Merrill Lynch. necessarily reflect the position of the Federal Reserve Bank of New York or Correspondence: [email protected] theFederalReserveSystem.
FRBNY Economic Policy Review / December 2013 1 Online Appendixes These appendixes, which depict graphically the processes described in the article, offer a comprehensive look at the shadow banking system and its many components.
Map: The Shadow Banking System www.newyorkfed.org/research/economists/adrian/1306adri_map.pdf
Appendix 1: The Government-Sponsored Shadow Banking System www.newyorkfed.org/research/economists/adrian/1306adri_A1.pdf
Appendix 2: The Credit Intermediation Process of Bank Holding Companies www.newyorkfed.org/research/economists/adrian/1306adri_A2.pdf
Appendix 3: The Credit Intermediation Process of Diversified Broker-Dealers www.newyorkfed.org/research/economists/adrian/1306adri_A3.pdf
Appendix 4: The Independent Specialists-Based Credit Intermediation Process www.newyorkfed.org/research/economists/adrian/1306adri_A4.pdf
Appendix 5: The Independent Specialists-Based Credit Intermediation Process www.newyorkfed.org/research/economists/adrian/1306adri_A5.pdf
Appendix 6: The Spectrum of Shadow Banks within a Spectrum of Shadow Credit Intermediation www.newyorkfed.org/research/economists/adrian/1306adri_A6.pdf
Appendix 7: The Pre-Crisis Backstop of the Shadow Credit Intermediation Process www.newyorkfed.org/research/economists/adrian/1306adri_A7.pdf
Appendix 8: The Post-Crisis Backstop of the Shadow Banking System www.newyorkfed.org/research/economists/adrian/1306adri_A8.pdf
The funding of credit through the shadow banking system governments have chosen to shield the traditional banking significantly reduced the cost of borrowing during the system from the risks inherent in maturity transformation by run-up to the financial crisis, at the expense of increasing the granting them access to backstop liquidity in the form of volatility of the cost of credit through the cycle. discount window lending and by providing a credit put to In particular, credit intermediaries’ reliance on short-term depositors in the form of deposit insurance. liabilities to fund illiquid long-term assets is an inherently In contrast to traditional banking’s public sector fragile activity that can make the shadow banking system prone guarantees, the shadow banking system, prior to the onset of to runs.2 During the financial crisis, the system came under the financial crisis, was presumed to be safe, owing to liquidity severe strain, and many parts of it collapsed. The emergence of backstops in the form of contingent lines of credit and tail-risk shadow banking thus shifted the systemic risk-return trade-off insurance in the form of wraps and guarantees. The credit lines toward cheaper credit intermediation during booms, at the and tail-risk insurance filled a backstop role for shadow banks cost of more severe crises and more expensive intermediation (much like the role discount window and deposit insurance during downturns. play for the commercial banking sector), but they were Shadow banks conduct credit, maturity, and liquidity provided by the private, not the public, sector. These forms of transformation much like traditional banks do. However, liquidity and credit insurance provided by the private sector, what distinguishes shadow banks is their lack of access to particularly commercial banks and insurance companies, public sources of liquidity, such as the Federal Reserve’s allowed shadow banks to perform credit, liquidity, and discount window, or to public sources of insurance, maturity transformation by issuing highly rated and liquid such as that provided by the Federal Deposit Insurance short-term liabilities. However, these guarantees also acted to Corporation (FDIC). Because the failure of credit transfer systemic risk between the core financial institutions intermediaries can have large, adverse effects on the real and the shadow banks. economy (see Bernanke [1983] and Ashcraft [2005]), As the solvency of the providers of private sector puts came into question (even if in some cases it was perfectly 2 Diamond and Dybvig (1983) initiated a large literature on bank runs modeled as multiple equilibria. Morris and Shin (2004) provide a model of funding satisfactory), the confidence that underpinned the stability of fragility with a unique equilibrium in a setting with higher-order beliefs. the shadow banking system vanished. The run on the system,
2Shadow Banking which began in the summer of 2007 and peaked following the government agencies during the financial crisis were direct failure of Lehman Brothers in September 2008, was curbed responses to the liquidity and capital shortfalls of shadow only after the creation of a series of official liquidity facilities banks. For example, the Commercial Paper Funding and credit guarantees that replaced private sector guarantees Facility (CPFF) provided emergency lending to issuers of entirely. In the interim, a large portion of the shadow banking commercial paper, the Primary Dealer Credit Facility supplied system collapsed, and several shadow intermediation activities a backstop for repo market borrowers, and the Term disappeared entirely. Asset-Backed Securities Loan Facility (TALF) offered ABS The assets and liabilities that collateralized and funded the to investors at “haircuts” below those available in times of shadow banking system were the product of a range of market distress. All three facilities directly provided liquidity securitization and secured lending techniques. Securitization support to shadow banking activities or entities, effectively refers to the pooling of mortgages, loans, receivables, and other offering a backstop for credit intermediation by the shadow financial cash flows into securities that are tranched according banking system and for traditional banks as a result of their to credit and liquidity characteristics. Secured lending refers to exposure to shadow banks. lending transactions that are secured by collateral, particularly Overviews of the shadow banking system are provided by securities, loan, or mortgage collateral. Pozsar (2008) and Adrian and Shin (2009). Pozsar catalogues Securitization-based credit intermediation potentially different types of shadow banks and describes the asset and increases the efficiency of credit intermediation. However, it funding flows within the shadow banking system. Adrian and also creates agency problems that do not exist when these Shin focus on the role of security brokers and dealers in the activities are conducted within a bank. Indeed, Ashcraft and shadow banking system, and discuss implications for financial Schuermann (2008) document seven agency problems that regulation. Our contribution with this article is to focus on arise in the securitization markets. If these agency problems are institutional details of the system, complementing a rapidly not adequately mitigated, the financial system is prone to growing literature on its collapse. As such, our study is excessive lowering of underwriting standards and to overly primarily descriptive and focuses on funding flows in a aggressive structuring of securities. somewhat mechanical manner. We believe that an The failure of private sector guarantees to support the understanding of the “plumbing” of the shadow banking shadow banking system occurred mainly because the relevant system is an important underpinning for any study of parties—credit rating agencies, risk managers, investors, and systemic interlinkages within the financial system. regulators—underestimated the aggregate risk and asset price The next section defines shadow banking and estimates its correlations. Specifically, the market did not correctly price for size. Section 3 discusses the seven steps of the shadow credit the fact that valuations of highly rated structured securities intermediation process. In section 4, we describe the become much more correlated in extreme environments than interaction of the shadow banking system with institutions during normal times. In a major systemic event, the price such as bank holding companies and broker-dealers. Section 5 behavior of diverse assets becomes highly correlated, as offers thoughts on the future of shadow banking. investors and levered institutions are forced to shed assets in order to generate the liquidity necessary to meet margin calls (see Coval, Jurek, and Stafford [2009]). 2. What Is Shadow Credit Correlations can also increase because of mark-to-market Intermediation? leverage constraints that result in “fire sales” (see Adrian and Shin [2010a] and Geanakoplos [2010]). The underestimation of correlations enabled financial institutions to hold 2.1 Defining Shadow Banking insufficient amounts of capital against the puts that underpinned the stability of the shadow banking system, which In the traditional banking system, credit intermediation made these puts unduly cheap to sell (see Gennaioli, Shleifer, between savers and borrowers occurs in a single entity. Savers and Vishny [forthcoming] for a model of the link between entrust their money to banks in the form of deposits, which the shadow banking and neglected risk). As investors also institutions use to fund the extension of loans to borrowers. overestimated the value of private credit and liquidity Banks furthermore issue debt and equity to capitalize their enhancement purchased through these puts, the result was an intermediation activities. Relative to direct lending (that is, excess supply of credit. In addition, the likely underpricing of savers lending directly to borrowers), banks issue safe, public sector liquidity and credit puts would have provided demandable deposits, thus removing the need for savers to further incentives for risk-taking behavior. monitor the risk-taking behavior of these institutions. The emergency liquidity facilities launched by the Federal Credit intermediation, the subset of financial Reserve and the guarantee schemes created by other intermediation that involves borrowing and lending through
FRBNY Economic Policy Review / December 2013 3 credit instruments, consists of credit, maturity, and liquidity sponsored enterprises (GSEs), which benefit from an transformation. Credit transformation refers to the implicit credit put to the taxpayer. enhancement of the credit quality of debt issued by the 3. Liabilities with indirect official enhancement generally intermediary through the use of priority of claims. For include the off-balance-sheet activities of depository insti- example, the credit quality of senior deposits is better than the tutions, such as unfunded credit card loan commitments credit quality of the underlying loan portfolio, owing to the and lines of credit to conduits. presence of more junior claims. Maturity transformation refers 4. Activities with indirect and implicit official enhancement to the use of short-term deposits to fund long-term loans, include asset management activities, such as bank- which creates liquidity for the saver but exposes the affiliated hedge funds and money market mutual funds intermediary to rollover and duration-mismatch risks. (MMMFs) as well as the securities lending activities of Liquidity transformation refers to the use of liquid instruments custodian banks. While financial intermediary liabilities to fund illiquid assets. For example, a pool of illiquid whole with an explicit enhancement benefit from official sector loans might trade at a lower price than a liquid rated security puts, liabilities enhanced with an implicit credit put option secured by the same loan pool, as certification by a credible might not benefit from such enhancements ex post. rating agency would reduce information asymmetries between Finally, some activities do not benefit from any form of borrowers and savers. official enhancement and are said to be unenhanced. An Credit intermediation is frequently enhanced through the example is guarantees made by monoline insurance use of third-party liquidity and credit guarantees, generally in companies. In addition, the securities lending activities of the form of liquidity or credit put options. A liquidity put insurance companies, pension funds, and certain asset option supplied by the private sector is typically provided in the managers do not benefit from access to official liquidity. form of contingent lines of credit by the commercial banking We define shadow credit intermediation to include all credit sector. Private sector credit put options are provided in the intermediation activities that are implicitly enhanced, form of wraps, guarantees, or credit default swaps (CDS) by indirectly enhanced, or unenhanced by official guarantees. insurance companies or banks. Liquidity and credit puts provided by the public sector consist of discount window access and deposit insurance. We call financial intermediation 2.2 Sizing the Shadow Banking System activities with public sector guarantees “officially enhanced.” Table 1 lays out the framework by which we analyze Before describing the shadow intermediation process in detail, official enhancements.3 Official enhancements to credit we provide a gauge for measuring the size of shadow banking intermediation can be classified into four categories, depending activity. The chart shows two measures of the shadow banking on whether they are direct or indirect and explicit or implicit. system, net and gross, both computed from the Federal Reserve’s “Flow of Funds” data. The gross measure sums all 1. A liability with direct official enhancement must reside on liabilities recorded in the Flow of Funds that relate to a financial institution’s balance sheet, while off-balance- securitization activity: mortgage-backed securities (MBS), ABS, sheet liabilities of financial institutions are indirectly and other GSE liabilities, as well as all short-term money market enhanced by the public sector. Activities with direct and explicit official enhancement include on-balance-sheet transactions that are not backstopped by deposit insurance: funding of depository institutions, insurance policies and repos, commercial paper, and other MMMF liabilities. The net annuity contracts, liabilities of most pension funds, and measure attempts to remove the double-counting. debt guaranteed through public sector lending programs.4 We should point out that these measures are imperfect for several reasons. First, the Flow of Funds data do not cover the 2. Activities with direct and implicit official enhancement include debt issued or guaranteed by the government- transactions of all shadow banking entities (see Eichner, Kohn, and Palumbo [2010] for data limitations of the Flow of Funds in detecting the imbalances that built up prior to the 3 A formal analysis of deposit insurance was conducted by Merton (1977) and Merton and Bodie (1993). financial crisis). 4 Depository institutions, including commercial banks, thrifts, credit unions, Second, we are not providing a measure of the shadow federal savings banks, and industrial loan companies, benefit from federal banks’ net supply of credit to the real economy. In fact, the deposit insurance and access to official liquidity backstops provided by the gross number sums up all shadow banking liabilities, discount window. Insurance companies benefit from guarantees provided by state guaranty associations. Defined-benefit private pensions benefit from irrespective of double-counting. The gross number should not insurance provided by the Pension Benefit Guaranty Corporation and public be interpreted as a proxy for the net supply of credit by shadow pensions benefit from implicit insurance provided by their state, municipal, or banks, but rather as the total balance-sheet capacity allocated to federal sponsors. The Small Business Administration, the U.S. Department of Education, and the Federal Housing Administration each operate programs shadow banking activities. The net measure mitigates the that provide explicit credit enhancement to private lending. second problem by netting the money market funding of ABS
4Shadow Banking Table 1 The Topology of Pre-Crisis Shadow Banking Activities and Liabilities Increasingly “Shadow” Credit Intermediation Activities ⎯→
Direct Public Enhancement Indirect Public Enhancement
Institution Explicit Implicit Explicit Implicit Unenhanced
Depository institutions Insured deposits Credit lines to Trust activities shadow banks Tri-party clearing Asset management Affiliate borrowing Commercial banks, clearing banks, ILCs Nondeposit liabilities
Federal loan programs DoE, SBA, and FHA credit puts Loan guarantees
Government-sponsored enterprises Fannie Mae, Freddie Mac, FHLBs Agency debt Agency MBS
Insurance companies Annuity liabilities Securities lending Insurance CDS protection sold policies
Pension funds Unfunded Securities lending liabilities
Diversified broker-dealers Brokered CP Tri-party repo MTNs Investment bank holding companies deposits (ILCs) Prime brokerage customer balances Liquidity puts (ABS, TOB, VRDO, ARS)
Mortgage insurers Financial guarantees Monoline insurers Financial guarantees CDS protection sold on CDOs Asset management (GICs, SIVs, conduits)
Shadow banks
Finance companies (stand-alones, captives) Brokered CP Team ABS, MTNs deposits (ILCs) ABCP Extendible ABCP Single-seller conduits ABCP Extendible ABCP Extendible ABCP Multiseller conduits ABCP Hybrid conduits ABCP Extendible ABCP Extendible ABCP TRS/repo conduits ABCP Securities arbitrage conduits ABCP Extendible ABCP Extendible ABCP Structured investment vehicles (SIVs) ABCP MTNs, capital notes Extendible ABCP Limited-purpose finance companies ABCP MTNs, capital notes Bilateral repo Bilateral repo Credit hedge funds (stand-alones) Bilateral repo Bilateral repo Money market intermediaries Shadow bank “depositors” Money market mutual funds $1 NAV Overnight sweep agreements $1 NAV Cash “plus” funds $1 NAV
FRBNY Economic Policy Review / December 2013 5 Table 1 (continued) The Topology of Pre-Crisis Shadow Banking Activities and Liabilities Increasingly “Shadow” Credit Intermediation Activities ⎯→
Direct Public Enhancement Indirect Public Enhancement
Institution Explicit Implicit Explicit Implicit Unenhanced
Enhanced cash funds $1 NAV Ultra-short bond funds $1 NAV Local government investment pools (LGIPs) $1 NAV Securities lenders $1 NAV
European banks Landesbanks, etc. State guarantees ABCP Credit lines to shadow banks
Source: Pozsar et al. (2012).
and MBS. As such, it is closer to a measure of the net supply of 3. The Shadow Credit credit provided by shadow banking activities, but it is still not Intermediation Process a perfect measure. Third, many of the securitized assets are held on the balance The shadow banking system is organized around securitization sheets of traditional depository and insurance institutions or and wholesale funding. Loans, leases, and mortgages are supported off their balance sheets through backup liquidity securitized and thus become tradable instruments. Funding is and credit derivative or reinsurance contracts. The holding of conducted in capital markets through instruments such as shadow liabilities by institutions inside the government safety commercial paper and repos. Savers hold money market net makes it difficult to draw bright lines between traditional balances instead of deposits with banks. and shadow credit intermediation, prompting us to classify the Like traditional banks, shadow banks conduct financial latter at the instrument level and not the institution level. intermediation. However, unlike in the traditional banking As shown in the chart on the next page, the gross measure of system, where credit intermediation is performed “under one shadow bank liabilities grew to nearly $22 trillion in June 2007. roof”—that of a bank—in the shadow banking system it is For comparison, we also plot total traditional banking performed through a chain of nonbank financial intermediaries liabilities, which were around $14 trillion in 2007.5 The size of in a multistep process. These steps entail the “vertical slicing” the shadow banking system has contracted substantially since of traditional banks’ credit intermediation process and include the peak in 2007, while total liabilities of the traditional 1) loan origination, 2) loan warehousing, 3) ABS issuance, banking sector have continued to grow throughout the crisis. 4) ABS warehousing, 5) ABS CDO issuance, 6) ABS The government’s liquidity facilities and guarantee “intermediation,” and 7) wholesale funding. The shadow schemes introduced in the summer of 2007 helped ease the banking system performs these steps of intermediation in a $5 trillion contraction in the size of the shadow banking strict, sequential order. Each step is handled by a specific type system, thereby protecting the broader economy from a of shadow bank and through a specific funding technique. collapse in the supply of credit as the financial crisis unfolded. Each of the seven steps of credit intermediation consists of a These programs were only temporary in nature; however, shadow banking activity, some of which is conducted by given the still-significant size of the shadow banking system specialized shadow banking institutions while others by and its exposure to runs by wholesale funding providers, one traditional financial intermediaries such as commercial banks open question is the extent to which some shadow banking or insurance companies. The seven steps of shadow bank activities should have more permanent access to official intermediation are as follows: backstops and receive more oversight. 1. Loan origination (such as auto loans and leases, noncon- forming mortgages) is performed by finance companies 5 Adrian and Shin (2010b) and Brunnermeier (2009) provide complementary that are funded through CP and medium-term notes overviews of the financial system in light of the financial crisis. (MTNs).
6Shadow Banking 2. Loan warehousing is conducted by single- and multi- Shadow Bank Liabilities versus Traditional seller conduits and is funded through asset-backed Bank Liabilities commercial paper (ABCP). Trillions of U.S. dollars 3. The pooling and structuring of loans into term asset- 25 backed securities are conducted by broker-dealers’ ABS Shadow liabilities syndicate desks. 20
4. ABS warehousing is facilitated through trading books and 15 is funded through repurchase agreements, total return swaps, or hybrid and repo conduits. 10 Bank liabilities 5. The pooling and structuring of ABS into CDOs are also 5 conducted by broker-dealers’ ABS syndicate desks. Net shadow liabilities 6. ABS intermediation is performed by limited-purpose 0 finance companies, structured investment vehicles (SIVs), 1210080604022000989694921990 securities arbitrage conduits, and credit hedge funds, which are funded in a variety of ways including, for Sources: Board of Governors of the Federal Reserve System, “Flow of example, repos, ABCP, MTNs, bonds, and capital notes. Funds Accounts of the United States” (as of 2011:Q3); Federal Reserve Bank of New York. 7. The funding of all of the above activities and entities is conducted in wholesale funding markets by funding providers such as regulated and unregulated money market intermediaries (for example, 2(a)-7 money market funds and enhanced cash funds, respectively) and direct money market investors (such as securities could be further repackaged into a CDO squared, which lenders). In addition to these cash investors—which fund would lengthen the intermediation chain to eight steps. shadow banks through short-term repo, CP, and ABCP Typically, the poorer an underlying loan pool’s quality at instuments—fixed-income mutual funds, pension funds, the beginning of the chain (for example, a pool of and insurance companies fund shadow banks by investing subprime mortgages originated in California in 2006), in their longer-term MTNs and bonds. the longer the credit intermediation chain will be to allow Shadow credit intermediation performs an economic role shadow credit intermediation to transform long-term, similar to that of traditional banks’ credit intermediation. The risky, and opaque assets into short-term and less risky shadow banking system decomposes the simple process of highly rated assets that can be used as collateral in short- retail-deposit-funded, hold-to-maturity lending conducted by term money markets. banks into a more complex, wholesale-funded, securitization- As a rule-of-thumb, the intermediation of low-quality based lending process. Through this intermediation process, long-term loans (nonconforming mortgages) involved all the shadow banking system transforms risky, long-term loans seven or more steps, whereas the intermediation of high- (subprime mortgages, for example) into seemingly credit-risk- quality short- to medium-term loans (credit card and auto free, short-term, money-like instruments, ending in wholesale loans) involved usually three steps and rarely more. The funding through stable net asset value shares issued by intermediation chain always starts with origination and 2(a)-7 MMMFs that require daily liquidity. This crucial point ends with wholesale funding, and each shadow bank is illustrated by the first and last links in the diagram, which appears only once in the process. depicts the asset and funding flows of the shadow banking system’s credit intermediation process. The intermediation 4. The Shadow Banking System steps of the shadow banking system are illustrated in Table 2. Importantly, not all intermediation chains involve all We identify three subgroups of the shadow banking seven steps, and some might involve even more. For system: 1) the government-sponsored shadow banking example, an intermediation chain might stop at step 2 if subsystem, 2) the “internal” shadow banking subsystem, a pool of prime auto loans is sold by a captive finance and 3) the “external” shadow banking subsystem. We company to a bank-sponsored multiseller conduit for term also discuss the liquidity backstops that were put in warehousing purposes. In another example, ABS CDOs place during the financial crisis.
FRBNY Economic Policy Review / December 2013 7 The Shadow Credit Intermediation Process
The shadow credit intermediation process consists of distinct steps. A credit intermediation chain, depending on the type and quality of credit involved, may entail as few as three steps and as many as seven or more. The shadow banking system conducts these steps in a string- sequential order. Each step is handled by specific types of financial entities, funded by specific types of liabilities.
“Asset flows”
Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7
Credit, maturity, Credit, maturity, Credit Credit, maturity, Credit Credit, maturity, Maturity and and liquidity and liquidity transformation and liquidity transformation and liquidity liquidity transformation transformation (blending) transformation (blending) transformation transformation
Loan Loan ABS ABS ABS CDO ABS Wholesale origination warehousing issuance warehousing issuance intermediation funding
Loans Loans Loans ABS ABS ABS CDO ABCP $1 NAV
CP ABCP Repo ABCP, repo CP, repo ABCP, repo
“Funding flows”
Source: Pozsar et al. (2012).
4.1 The Government-Sponsored Shadow funded not through deposits but through the capital markets, Banking Subsystem where they issue short- and long-term agency debt securities. These securities are bought by money market investors and The seeds of the shadow banking system were sown nearly real-money investors such as fixed-income mutual funds. The eighty years ago with the creation of government-sponsored funding functions performed by the GSEs on behalf of banks enterprises, which include the Federal Home Loan Bank and the way GSEs are funded are the models for wholesale (FHLB) system in 1932, the Federal National Mortgage funding markets (see Table 3 and Online Appendix 1). Association (Fannie Mae) in 1938, the Government National The GSEs have embodied five intermediation techniques: Mortgage Association (Ginnie Mae) in 1968, and the Federal Home Loan Mortgage Corporation (Freddie Mac) in 1970. 1. term loan warehousing provided to banks by the FHLBs, Each of these institutions is perceived by the marketplace to be 2. credit risk transfer and transformation through credit a shadow bank given that its liabilities are implicitly guaranteed insurance provided by the GSEs, by U.S. taxpayers. The GSEs have had a large influence on the 3. originate-to-distribute securitization functions provided way in which the financial system is funded and conducts credit for banks by the GSEs, transformation. Arguably, they were the first providers of term warehousing of loans and invented the originate-to-distribute 4. maturity transformation conducted through the GSE- retained portfolios, and model of securitized credit intermediation. GSEs largely securitize their loan and mortgage portfolios in 5. pass-through MBS funding of mortgage credit. pools of mortgage-backed securities, which are referred to as Over the past thirty years, these techniques were developed agency MBS. These MBS pass interest payments and principal by dealers, banks, and the GSEs and became the foundation for payments through to the MBS holder, but the credit risk is the securitization process of shadow credit intermediation. The retained by the GSEs. Agency MBS thus incorporate interest adaptation of these techniques fundamentally changed the rate and prepayment risk, but not the default risk of individual bank-based, originate-to-hold credit intermediation process borrowers. Freddie Mac issued the first pass-through certificate and gave rise to the securitization-based, originate-to- in 1971, while the first pass-through MBS were issued by distribute credit intermediation process. Ginnie Mae in 1970 and Fannie Mae in 1981. The government-sponsored shadow banking subsystem is The MBS that are retained by the GSEs are funded with a not involved in loan origination, only in loan processing and maturity mismatch. Unlike banks, however, the GSEs are funding.6 These entities qualify as shadow banks to the extent
8Shadow Banking Table 2 Examples of the Steps, Entities, and Funding Techniques of the Shadow Credit Intermediation Process
Step Function Shadow Banks Shadow Banks’ Funding Techniques
1 Loan origination Finance companies CP, MTNs, bonds 2 Loan warehousing Single- and multiseller conduits ABCP 3 ABS issuance SPVs, structured by broker-dealers ABS 4ABS warehousingHybrid, TRS/repo conduits, broker-dealers’ trading books ABCP, repo 5 ABS CDO issuance SPVs, structured by broker-dealers ABS CDOs, CDO-squareds 6 ABS intermediation LPFCs, SIVs, securities arbitrage conduits, credit hedge funds ABCP, MTNs, repo 7 Wholesale funding 2(a)-7 MMMFs, enhanced cash funds, securities lenders, etc. $1 NAV shares (shadow bank “deposits”)
Source: Pozsar et al. (2012).
Notes: Entries in bold denote securitized funding techniques. Securitized funding techniques are not synonymous with secured funding.
Table 3 Examples of the Steps, Entities, and Funding Techniques of the GSE Credit Intermediation Process
Step Function Shadow Banks Shadow Banks’ Funding Techniques 1 Mortgage origination Commercial banks Deposits, CP, MTNs, bonds 2 Mortgage warehousing FHLBs Agency debt, discount notes 3 ABS issuance Fannie Mae, Freddie Mac through TBA market Agency MBS (pass-through) 4 ABS warehousing Broker-dealers’ trading books ABCP, repo 5 ABS CDO issuance Broker-dealer agency MBS desks CMOs (resecuritizations) 6 ABS intermediation GSE retained portfolios Agency debt, discount notes 7 Wholesale funding 2(a)-7 MMMFs, enhanced cash funds, securities lenders $1 NAV shares (GSE “deposits”)
Source: Pozsar et al. (2012).
Notes: Entries in bold denote securitized funding techniques. Securitized funding techniques are not synonymous with secured funding.
that they are involved in the traditional bank activities of credit, profitability. The largest banks and dealers played a central role maturity, and liquidity transformation, but without actually in the development of shadow banking activities, particularly in being chartered as banks and without having meaningful access the origination, warehousing, securitizing, and funding of credit. to a lender of last resort and an explicit insurance of their As a result, the nature of banking changed from a credit-risk- liabilities by the federal government.7 intensive, deposit-funded, spread-based business to a less credit- risk-intensive, wholesale-funded process subject to run risk. 4.2 The “Internal” Shadow The vertical and horizontal slicing of credit intermediation uses a range of off-balance-sheet securitization and asset Banking Subsystem management techniques, which enable banks to conduct lending with less capital than if they had retained loans on their The development of the GSEs’ activities has been mirrored by the balance sheets (Table 4). This process enhances the RoE of development of a full-fledged shadow banking system. In recent banks—or, more precisely, the RoE of their holding companies. decades, the largest banks were transformed from low- Shadow banking activities of bank holding companies return-on-equity (RoE) utilities, originating loans and holding (BHCs) are conducted off balance sheet through various and funding them until maturity, to high-RoE entities that subsidiaries. BHCs: 1) originate loans in their bank or finance developed shadow banking activities in order to increase company subsidiaries, 2) warehouse and accumulate loans in 6 By design, GSEs are prohibited from loan origination. They create a off-balance-sheet conduits that are managed by their broker- secondary market for mortgages to facilitate their funding. dealer subsidiaries, with funding through wholesale funding 7 Note that Fannie Mae and Freddie Mac had some explicit backstops from the markets and liquidity enhancements by bank subsidiaries, U.S. Treasury in the form of credit lines prior to their conservatorship in 2008. However, these liquidity backstops were very small compared with the size of 3) securitize loans through their broker-dealer subsidiaries by the agencies’ balance sheets. transferring them from the conduit into bankruptcy-remote
FRBNY Economic Policy Review / December 2013 9 Table 4 Examples of the Steps, Entities, and Funding Techniques of the FHC Credit Intermediation Process
Step Function Shadow Banks Shadow Banks’ Funding Techniques
1 Loan origination Commercial bank subsidiary Deposits, CP, MTNs, bonds 2 Loan warehousing Single- and multi-seller conduits ABCP 3 ABS issuance SPVs, structured by broker-dealer subsidiary ABS 4 ABS warehousing Hybrid, TRS/repo conduits, broker-dealers’ trading books ABCP, repo 5 ABS CDO issuance SPVs, structured by broker-dealer subsidiary ABS CDOs, CDO-squareds 6 ABS intermediation SIVs, internal credit hedge funds (asset management) ABCP, MTNs, capital notes, repo 7 Wholesale funding 2(a)-7 MMMFs, enhanced cash funds, securities lending subsidiary $1 NAV shares (shadow bank “deposits”)
Source: Pozsar et al. (2012). Notes: Entries in bold denote securitized funding techniques. Securitized funding techniques are not synonymous with secured funding.
special-purpose vehicles, and 4) fund the safest tranches of This interpretation of the workings of BHCs is different structured credit assets in off-balance-sheet ABS intermediaries from the one emphasizing the benefits of BHCs as “financial (such as SIVs) that are managed from the asset management supermarkets.” According to that widely held view, the subsidiary of the holding company and are funded through diversification of the holding companies’ revenues through wholesale funding markets with backstops by the bank broker-dealer and asset management activities makes the subsidiaries (see Online Appendix 2). banking business more stable, as the holding companies’ This process highlights three important aspects of the banks, if need be, could be supported by net income from other changed nature of lending in the U.S. financial system, operations during times of credit loss. In our interpretation, especially for residential and commercial mortgage credit. the broker-dealer and asset management activities are not First, the process of lending and the uninterrupted flow of parallel, but instead are serial and complementary activities to credit to the real economy no longer rely only on banks, but on BHCs’ banking activities. a process that spans a network of banks, broker-dealers, asset managers, and shadow banks funded through wholesale funding and capital markets globally. Second, bank subsidiaries’ only direct involvement in the 4.3 The “External” Shadow Banking shadow credit intermediation process is at the loan origination level. The indirect involvement of commercial bank subsidiaries is Subsystem broader, however, as the banks act as lenders to other subsidiaries and off-balance-sheet vehicles involved in the warehousing and processing of loans, as well as the distribution and funding of Similar to the “internal” shadow banking subsystem, the structured credit securities. Even though a BHC’s credit “external” version is a global network of balance sheets. The intermediation process depends on at least four entities other than origination, warehousing, and securitization of loans are the bank, only the bank subsidiary of a BHC has access to the Federal conducted mainly from the United States, but the funding and Reserve’s discount window and the benefits of deposit insurance. maturity transformation of structured credit assets are Third, securitization techniques have increased the implicit conducted from the United States, Europe, and offshore leverage of bank holding companies, sometimes called “capital financial centers. While the internal subsystem is designed efficiency.” As the financial crisis of 2007-09 showed, however, the primarily to raise the profitability of BHCs by increasing their capital efficiency of the process is highly dependent on liquid effective leverage through off-balance-sheet entities and wholesale funding and debt capital markets globally. The exposure activities, the external subsystem has resulted from vertical of BHCs to shadow bank entities increases the effective leverage of integration and the exploitation of gains from specialization. the BHC, even though that might not be obvious from looking at The external shadow banking subsystem is defined by the balance sheet because much shadow banking activity is designed 1) the credit intermediation process of diversified broker- to be conducted off balance sheet. The implicit leverage in turn dealers, 2) the credit intermediation process of independent, exposes BHCs to credit and liquidity risk and represents an nonbank specialist intermediaries, and 3) the credit puts important source of systemic risk. provided by private credit-risk repositories.
10 Shadow Banking Table 5 Examples of the Steps, Entities, and Funding Techniques of the DBD Credit Intermediation Process
Step Function Shadow Banks Shadow Banks’ Funding Techniques
1 Loan origination Finance company subsidiary CP, MTNs, bonds 2 Loan warehousing Single- and multi-seller conduits ABCP 3 ABS issuance SPVs, structured by broker-dealer subsidiary ABS 4 ABS warehousing Hybrid, TRS/repo conduits, broker-dealers’ trading books ABCP, repo 5 ABS CDO issuance SPVs, structured by broker-dealer subsidiary ABS CDOs, CDO-squareds 6 ABS intermediation Internal credit hedge funds, proprietary trading desks Repo 7 Wholesale funding 2(a)-7 MMMFs, enhanced cash funds, securities lending subsidiary $1 NAV shares (shadow bank “deposits”)
Source: Pozsar et al. (2012).
Notes: Entries in bold denote securitized funding techniques. Securitized funding techniques are not synonymous with secured funding.
The Credit-Intermediation Process of Diversified The Independent-Specialists-Based Credit Broker-Dealers Intermediation Process
We refer to the stand-alone investment banks as they existed The credit intermediation process that runs through a network prior to 2008 as diversified broker-dealers (DBDs). DBDs of independent specialists is the same as those of FHCs and vertically integrate their securitization businesses (from DBDs and results in the same credit intermediation functions origination to funding), lending platforms, and asset as those performed by traditional banks. The independent- management units. The credit intermediation process of DBDs specialists-based intermediation process includes the following is similar to that of financial holding companies (FHCs; Table 5). types of entities: stand-alone and captive finance companies on 8 The diversified broker-dealers are distinguished by the fact the loan origination side; independent multiseller conduits on that they do not own commercial bank subsidiaries. Some of the loan warehousing side; and limited-purpose finance the major stand-alone investment banks did, however, own companies, independent SIVs, and credit hedge funds on the industrial loan company (ILC) subsidiaries. However, owning ABS intermediation side (Table 6). an ILC did not require the holding company to turn into a bank There are three key differences between the independent- holding company. Since running one’s own loan warehouses specialists-based credit intermediation process and those of (single- or multiseller loan conduits) requires large bank BHCs and DBDs. First and foremost, on the origination side, subsidiaries to fund the contingent liquidity backstops that the three processes intermediate different types of credit. The enhance the ABCP issued by the conduits, broker-dealers BHC and DBD processes originate some combination of both typically outsourced these warehousing functions to BHCs conforming and nonconforming mortgages, as well as with large deposit bases or to independent multiseller conduits. commercial mortgages, leveraged loans, and credit card loans. At the end of their intermediation chains, DBDs do not In contrast, the independent-specialists-based process tends to operate securities arbitrage conduits or SIVs. Instead, the specialize in the origination of auto and equipment loans and dealers run internal credit hedge funds, fund trading books, leases, middle-market loans, franchise loans, and more esoteric and fund repo conduits. The intermediation process of DBDs loans. The obvious exceptions to this are stand-alone tends to rely more on repo funding than that of FHCs, which nonconforming mortgage finance companies, which have rely on CP, ABCP, MTNs, and repos. The subsidiaries of DBDs become largely extinct since the crisis. do not have direct access to public sources of credit or liquidity The independent-specialists-based credit intermediation process backstops. It should be noted that the credit intermediation is based on an “originate-to-fund” model (again, with the exception processes described here are the simplest and shortest forms of of the now extinct stand-alone mortgage finance companies), as the intermediation chains that run through FHCs and DBDs. opposed to the mostly originate-to-distribute model of the In practice, these processes are often elongated by additional 8 Captive finance companies are finance companies that are owned by steps involved in the warehousing, processing, and distribution nonfinancial corporations, typically manufacturing firms or homebuilders. They are used to provide vendor financing to the clients of their parents and of unsold ABS into ABS CDOs (see Online Appendix 3). benefit from cross-guarantees. Stand-alone finance companies, as the name suggests, stand on their own and are not subsidiaries of any corporate entity.
FRBNY Economic Policy Review / December 2013 11 Table 6 Examples of the Steps, Entities, and Funding Techniques of the Independent-Specialists-Based Credit Intermediation Process
Step Function Shadow Banks Shadow Banks’ Funding Techniques
1 Loan origination Stand-alone and captive finance companies CP, MTNs, bonds 2 Loan warehousing FHC-sponsored and independent multiseller conduits ABCP 3 ABS issuance SPVs, structured by broker-dealers ABS 4 ABS warehousing ABCP, repo 5 ABS CDO issuance ABS CDOs, CDO-squareds 6 ABS intermediation LPFCs, independent SIVs, independent credit hedge funds ABCP, MTNs, capital notes, repo 7 Wholesale funding 2(a)-7 MMMFs, enhanced cash funds, securities lenders $1 NAV shares (shadow bank “deposits”)
Source: Pozsar et al. (2012).
Notes: Entries in bold denote securitized funding techniques. Securitized funding techniques are not synonymous with secured funding.
government-sponsored shadow banking subsystem and the credit shadow bank intermediation chain (step 5). By providing such intermediation processes of BHCs and DBDs. tail-risk insurance, the private credit-risk repositories change While the GSE, BHC, and DBD credit intermediation the pricing of tail risk and ultimately affect the supply of credit processes depend heavily on liquid capital markets for their to the real economy. ability to fund, securitize, and distribute loans, independent Different credit-risk repositories correspond to specific specialists’ seamless functioning is also exposed to the ability of stages of the shadow credit intermediation process. As such, DBDs and FHCs to perform their functions as gatekeepers to mortgage insurers specialize in insuring or wrapping whole capital markets and lenders of last resort, respectively. This in mortgage loans; monoline insurers, which are bond insurance turn represents an extra layer of fragility in the structure of the companies, specialize in wrapping ABS tranches (or the loans independent-specialists-based credit intermediation process, backing specific ABS tranches), and diversified insurance as failure by FHCs and DBDs to perform these functions in companies and credit hedge funds take on the risks of ABS times of systemic stress runs the risk of paralyzing and disabling CDO tranches through CDS.9 the process (see Rajan [2005]). Effectively, the various forms of credit put options provided Indeed, this fragility became apparent during the financial by private risk repositories absorb tail risk from loan pools, crisis of 2007-09, as the independent-specialists-based process turning the enhanced securities into less risky ones (at least from broke down and with it the flow of corresponding types of the perspective of investors prior to the crisis). This in turn credit to the real economy. Online Appendix 4 describes the means that any liability issued against these assets is perceived to relative extent to which specialist loan originators (captive and be less risky as well. Such credit puts provided by risk repositories independent finance companies) relied on BHCs and DBDs as to the shadow banking system thus play a role analogous to FDIC their ABS underwriters and gatekeepers to capital markets. insurance for the commercial banking system. The perceived credit-risk-free nature of traditional bank and shadow bank liabilities stems from two very different Private Credit-Risk Repositories sources. In the case of traditional banks’ insured liabilities The shadow credit intermediation processes of independent (deposits), the credit quality is driven by the counterparty: the specialists, BHCs, and DBDs rely heavily on private credit-risk U.S. taxpayer. As a result, insured depositors do not need to repositories (see Online Appendix 5). Private risk repositories examine a bank’s creditworthiness before depositing money— specialize in providing credit transformation services in the it is the regulator that performs the due diligence. In the case of shadow banking system, and include mortgage insurers, shadow bank liabilities such as repos or ABCP, perceived credit monoline insurers, diversified insurance companies, and credit 9 CDS were also used for hedging warehouse and counterparty exposures. hedge funds. These entities facilitate the securitization process For example, a broker-dealer with a large exposure to subprime MBS that by providing tail-risk insurance for structured credit products it warehoused for an ABS CDO deal in the making could purchase CDS in various forms. For example, insurance companies might protection on its MBS warehouse. In turn, the broker-dealer could also purchase protection (a counterparty hedge) from a credit hedge fund or offer CDS written on mezzanine tranches of ABS, thus credit derivative product company on the counterparty providing the CDS enhancing credit ratings at the resecuritization stage of the protection on subprime MBS.
12 Shadow Banking quality is based on the credit enhancements achieved through 4.5 Backstopping the Shadow Banking private credit-risk repositories. Credit rating agencies, in turn, System perform the due diligence on behalf of the ultimate investors. The credit puts of private credit-risk repositories also The Federal Reserve’s 13(3) emergency lending facilities that perform a function similar to that of the wraps provided by followed the bankruptcy of Lehman Brothers amount to a Fannie Mae and Freddie Mac on conforming mortgage pools, backstop for all the functional steps involved in the shadow as these government-sponsored, public credit-risk repositories credit intermediation process. The facilities introduced during allow senior mortgage tranches to achieve AAA ratings by the crisis were an explicit recognition of the need to channel 10 removing credit risk. emergency funds into internal, external, and government- sponsored shadow banking subsystems. (To read about a pre- and postcrisis backstop for shadow banks, see Online 4.4 The “Parallel” Banking System Appendixes 7 and 8.) As such, the CPFF was a backstop for the CP and ABCP Many “internal” and “external” shadow banks existed in a form issuance of loan originators and loan warehouses, respectively that was possible only because of the special circumstances in (steps 1 and 2 of the shadow credit intermediation process); the the run-up to the financial crisis. Some of these circumstances TALF was a backstop for ABS issuance (step 3); Maiden Lane were economic in nature and some were due to regulatory and LLC was a backstop for Bear Stearns’ ABS warehouse, while the risk management failures. However, there are also examples of Term Securities Lending Facility (TSLF) was a means to shadow banks that had competitive advantages relative to improve the average quality of broker-dealers’ securities traditional banks. These shadow banks were driven not by warehouses by swapping ABS for Treasury securities (step 4); regulatory arbitrage, but by gains from specialization as a Maiden Lane III LLC was a backstop for AIG-Financial “parallel” banking system. Most of these entities were found in Products’ credit puts on ABS CDOs (step 5); and the Term the “external” shadow banking subsystem. Auction Facility (TAF) and foreign exchange swaps with other These entities include nonbank finance companies, which central banks were meant to facilitate the “onboarding” and can be more efficient than traditional banks because of on-balance-sheet dollar funding of the ABS portfolios of specialization and economies of scale in the origination, formerly off-balance-sheet ABS intermediaries—mainly SIVs servicing, structuring, trading, and funding of loans to both and securities arbitrage conduits (step 6).12 11 bankable and nonbankable credits. For example, finance The Primary Dealer Credit Facility (PDCF) was a backstop companies have traditionally served subprime credit card or for the funding of diversified broker-dealers through the tri- auto loan customers, as well as low-rated corporate credits party repo system. The Asset-Backed Commercial Paper such as the commercial airlines, none of which are served by Money Market Mutual Fund Liquidity Facility (AMLF) and the banks. Furthermore, some ABS intermediaries could fund Money Market Investor Funding Facility (MMIFF) served as highly rated structured credit assets at a lower cost and at liquidity backstops for regulated and unregulated money lower levels of leverage than banks could with high-return- market intermediaries, respectively (step 7). The FDIC’s on-equity targets. Temporary Liquidity Guarantee Program, which covered Over the last thirty years, a number of activities have been various bank and nonbank financial institutions’ senior pushed out of banks and into the parallel banking system. It unsecured debt and corporations’ non-interest-bearing remains an open question whether the parallel banking system deposit transaction accounts, regardless of dollar amount, will ever remain stable through credit cycles in the absence of was another emergency backstop, as was the U.S. Treasury official credit and liquidity puts. If the answer is no, then there Department’s temporary guarantee program of retail and are questions about whether such puts and the associated institutional money market mutual funds. prudential controls should be extended to parallel banks or, Upon the complete rollout of the liquidity facilities and alternatively, whether parallel banking activity should be guarantee schemes, the shadow banking system was fully severely restricted. (A spectrum of shadow banking activities by embraced by official credit and liquidity puts and became fully type is described in Online Appendix 6.) backstopped, just like the traditional banking system. As a result, the adverse effect on real economic activity from the collapse of the shadow banking system was mitigated.
12 10 The CPFF is documented in detail by Adrian, Marchioni, and Kimbrough Credit wraps come in different forms and guarantee the timely payment of (2011); the TSLF is described by Fleming, Hrung, and Keane (2009); the TALF principal and interest on an underlying debt obligation. is described by Campbell et al. (2011) and Ashcraft, Malz, and Pozsar (2012); 11 Carey, Post, and Sharpe (1998) document the specialization of finance the PDCF is discussed by Adrian, Burke, and McAndrews (2009); the TAF is companies and their servicing of riskier borrowers. documented by Armantier, Krieger, and McAndrews (2008).
FRBNY Economic Policy Review / December 2013 13 5.Conclusion on- and off-balance-sheet activities—all under the umbrella of financial holding companies. Finally, the “external” shadow We document the specialized financial institutions of the banking subsystem refers to the credit intermediation process shadow banking system and argue that these credit of diversified broker-dealers and a global network of intermediaries played a quantitatively important role in the independent, nonbank financial specialists that includes run-up to the financial crisis. Shadow credit intermediation captive and stand-alone finance companies, limited-purpose includes three broad types of activities differentiated by their finance companies, and asset managers. strength of official enhancement: implicitly enhanced, While much of the current and future reform efforts are indirectly enhanced, and unenhanced. focused on remediating the excesses of the recent credit The shadow banking system has three subsystems that bubble, we note that increased capital and liquidity standards intermediate different types of credit in fundamentally for depository institutions and insurance companies are likely different ways. The government-sponsored shadow banking to increase the returns to shadow banking activity. For subsystem refers to credit intermediation activities funded example, as pointed out in Pozsar (2011), the reform effort through the sale of agency debt and MBS, which mainly include has done little to address the tendency of large institutional conforming residential and commercial mortgages. The cash pools to form outside the banking system. Thus, we “internal” shadow banking subsystem refers to the credit expect shadow banking to be a significant part of the financial intermediation process of a global network of banks, finance system, although almost certainly in a different form, for the companies, broker-dealers, and asset managers and their foreseeable future.
14 Shadow Banking References
Adrian, T., C. Burke, and J. McAndrews. 2009. “The Federal Reserve’s Campbell, S., D. Covitz, W. Nelson, and K. Pence. 2011. “Securitization Primary Dealer Credit Facility.” Federal Reserve Bank New York Markets and Central Banking: An Evaluation of the Term Asset- Current Issues in Economics and Finance 15, no. 4 (August). Backed Securities Loan Facility.” Journal of Monetary Economics 58, no. 5 (July): 518-31. Adrian, T., D. Marchioni, and K. Kimbrough. 2011. “The Federal Reserve’s Commercial Paper Funding Facility.” Federal Reserve Bank Carey, M., M. Post, and S. A. Sharpe. 1998. “Does Corporate of New York Economic Policy Review 17, no. 1 (May): 25-39. Lending by Banks and Finance Companies Differ? Evidence on Specialization in Private Debt Contracting.” Journal of Adrian, T., and H. S. Shin. 2009. “The Shadow Banking System: Finance 53, no. 3 (June): 845-78. Implications for Financial Regulation.” Banque de France Financial Stability Review, no. 13 (September): 1-10. Coval, J., J. Jurek, and E. Stafford. 2009. “The Economics of Structured Finance.” Journal of Economic Perspectives 23, ———. 2010a. “Liquidity and Leverage.” Journal of Financial no. 1 (winter): 3-25. Intermediation 19, no. 3 (July): 418-37. Covitz, D., N. Liang, and G. Suarez. 2009. “The Evolution of a ———. 2010b. “The Changing Nature of Financial Intermediation Financial Crisis: Panic in the Asset-Backed Commercial Paper and the Financial Crisis of 2007-2009.” Annual Review of Market.” Board of Governors of the Federal Reserve System Economics 2, September: 603-18. Finance and Economics Discussion Series, no. 2009-36.
Aitken, J., and M. Singh. 2009. “Deleveraging after Lehman: Diamond, D., and P. Dybvig. 1983. “Bank Runs, Deposit Insurance, Evidence from Reduced Rehypothecation.” IMF Working and Liquidity.” Journal of Political Economy 91, no. 3 Paper no. 09/42, July. (June): 401-19.
Armantier, O., S. Krieger, and J. McAndrews. 2008. “The Federal Eichner, M., D. Kohn, and M. Palumbo. 2010. “Financial Statistics for Reserve’s Term Auction Facility.” Federal Reserve Bank of New York the United States and the Crisis: What Did They Get Right, What Current Issues in Economics and Finance 14, no. 5 (July). Did They Miss, and How Should They Change?” Board of Governors of the Federal Reserve System Finance and Ashcraft, A. B. 2005. “Are Banks Really Special? New Evidence from Economics Discussion Series, no. 2010-20. the FDIC-Induced Failure of Healthy Banks.” American Economic Review 95, no. 5 (December): 1712-30. Fleming, M., W. Hrung, and F. Keane. 2009. “The Term Securities Lending Facility: Origin, Design, and Effects.” Federal Reserve Ashcraft, A. B., A. Malz, and Z. Pozsar. 2012. “The Federal Reserve’s Bank of New York Current Issues in Economics and Term Asset-Backed Securities Loan Facility.” Federal Reserve Bank of Finance 15, no. 2 (February). New York Economic Policy Review 18, no. 3 (November): 29-66. Geanakoplos, J. 2010. “The Leverage Cycle.” In Daron Acemoglu, Ashcraft, A. B., and T. Schuermann. 2008. “Understanding the Kenneth Rogoff, and Michael Woodford, eds., NBER Securitization of Subprime Mortgage Credit.” Foundations Macroeconomics Annual 2009. Chicago: University and Trends in Finance 2, no. 3 (July): 191-309. of Chicago Press.
Bernanke, B. S. 1983. “Nonmonetary Effects of the Financial Crisis in Gennaioli, N., A. Shleifer, and R. Vishny. Forthcoming. “A Model of the Propagation of the Great Depression.” American Economic Shadow Banking.” Journal of Finance. Review 73, no. 3 (June): 257-76. Gorton, G. 2009. “Slapped in the Face by the Invisible Hand: Banking Brunnermeier, M. 2009. “Deciphering the Liquidity and Credit and the Panic of 2007.” Paper presented at the Federal Reserve Crunch 2007-2008.” Journal of Economic Perspectives 23, Bank of Atlanta’s 2009 Financial Markets Conference on Financial no. 1 (winter): 77-100. Innovation and Crisis, Atlanta, Georgia, May 11-13.
FRBNY Economic Policy Review / December 2013 15 References (Continued)
Martin, A., D. Skeie, and E. von Thadden. 2010. “Repo Runs.” Federal Pozsar, Z. 2008. “The Rise and Fall of the Shadow Banking System.” Reserve Bank of New York Staff Reports, no. 444, April. Moody’s economy.com.
McCulley, P. 2007. “Teton Reflections.” PIMCO Global Central ———. 2011. “Institutional Cash Pools and the Triffin Dilemma of Bank Focus, September. the U.S. Banking System.” IMF Working Paper no. 11/190, August.
Merton, R. C. 1977. “An Analytical Derivation of the Cost of Deposit Pozsar, Z., T. Adrian, A. Ashcraft, and H. Boesky. 2012. “Shadow Insurance and Loan Guarantees: An Application of Modern Banking.” Revue d’Economie Financière 105: 157-84. Option Pricing Theory.” Journal of Banking and Finance 1, no. 1 (June): 3-11. Rajan, R. 2005. “Has Financial Development Made the World Riskier?” Paper presented at the Federal Reserve Bank of Kansas City Merton, R. C., and Z. Bodie. 1993. “Deposit Insurance Reform: Symposium on The Greenspan Era: Lessons for the Future, A Functional Approach.” Carnegie-Rochester Conference Jackson Hole, Wyoming, August 25-27. Series on Public Policy 38, June: 1-34. Wachter, S., A. Pavlov, and Z. Pozsar. 2008. “Subprime Lending and Morris, S., and H. S. Shin. 2004. “Coordination Risk and the Price of Real Estate Markets.” In Stefania Perrucci, ed., Mortgage and Debt.” European Economic Review 48, no. 1 (February): 133-53. Real Estate Finance. London: Risk Books.
The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.
16 Shadow Banking Andreas Fuster, Laurie Goodman, David Lucca, Laurel Madar, Linsey Molloy, and Paul Willen
The Rising Gap between Primary and Secondary Mortgage Rates
• While the primary-secondary mortgage 1. Introduction rate spread is a closely tracked series, it is an imperfect measure of the pass-through he vast majority of mortgage loans in the United States between secondary-market valuations and Tare securitized in the form of agency mortgage- primary-market borrowing costs. backed securities (MBS). Principal and interest payments on these securities are passed through to investors and • This study tracks cash flows during and after are guaranteed by the government-sponsored enterprises the mortgage origination and securitization (GSEs) Fannie Mae or Freddie Mac or by the government process to determine how many dollars organization Ginnie Mae.1 Thus, investors in these securities (per $100 loan) are absorbed by originators, are not subject to loan-specific credit risk; they face only either to cover costs or as originator profits. interest rate and prepayment risk—the risk that borrowers may refinance the loan when rates are low.2 • The authors calculate a series of originator In the primary mortgage market, lenders make loans to profits and unmeasured costs (OPUCs) for borrowers at a certain interest rate, whereas in the secondary the period 1994-2012, and show that these market, lenders securitize these loans into MBS and sell them OPUCs increased significantly between to investors. When thinking about the relationship between 2008 and 2012. these two markets, policymakers and market commentators usually pay close attention to the “primary-secondary spread.” • Although some mortgage origination costs This spread is calculated as the difference between an average may have risen, a large component of the rise in OPUCs remains unexplained by 1 Fannie Mae is the Federal National Mortgage Association (or FNMA); cost increases alone, pointing to increased Freddie Mac is the Federal Home Loan Mortgage Corporation (FHLMC; profitability of originators. also FGLMC); Ginnie Mae is the Government National Mortgage Association (GNMA). 2 They also face the risk that borrowers prepay at lower-than-expected speeds when interest rates rise.
Andreas Fuster and David Lucca are senior economists in the Federal Reserve This article is a revised version of a white paper originally prepared as back- Bank of New York’s Research and Statistics Group; Laurie Goodman is the ground material for the workshop “The Spread between Primary and Secondary center director of the Housing Finance Policy Center at the Urban Institute; Mortgage Rates: Recent Trends and Prospects,” held at the Federal Reserve Laurel Madar and Linsey Molloy are associates in the Bank’s Markets Group; Bank of New York on December 3, 2012. The authors thank Adam Ashcraft, Paul Willen is a senior economist and policy advisor in the Federal Reserve Alan Boyce, James Egelhof, David Finkelstein, Kenneth Garbade, Brian Landy, Bank of Boston’s Research Department. Jamie McAndrews, Joseph Tracy, and Nate Wuerffel for helpful comments, Corresponding authors: [email protected]; [email protected] and Shumin Li for help with the data. The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York, the Federal Reserve Bank of Boston, or the Federal Reserve System. FRBNY Economic Policy Review / December 2013 17 Chart 1 C The Primary-Secondary Spread Back-of-the-Envelope Calculation of the Net Market Value of a Thirty-Year Fixed-Rate Basis points Mortgage Securitized in an Agency MBS 175
150 Dollars per $100 loan 5 125 Weekly Eight-week 4 100 rolling window 75 3
50 2 25 1 0 1995 96 98 00 02 04 06 08 10 12 0 2006 07 08 09 10 11 12 Sources: Bloomberg L.P.; Freddie Mac.
Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; authors’ calculations. Notes: e chart shows the interpolated value of a mortgage-backed mortgage interest rate (usually coming from the Freddie Mac security (MBS) with coupon (rprimary – g-fee) minus 100. e line reects an eight-week rolling window average; the calculation uses Primary Mortgage Market Survey) and a representative yield back-month MBS prices. on newly issued agency MBS—the “current-coupon rate.” Chart 1 shows a time series of the primary-secondary spread through the end of 2012. The spread was relatively stable from 1995 to 2000, at about 30 basis points; it To get a sense of what lenders earn from selling loans, we subsequently widened to about 50 basis points through early first consider a simple “back-of-the-envelope” calculation. 2008, but then reached more than 100 basis points in early We track the secondary-market value of the typical offered 2009 and during 2012. Following the September 2012 Federal mortgage loan (according to the Freddie Mac survey) over Open Market Committee announcement of additional MBS time, assuming that the lender securitizes and sells the loan purchases, the spread temporarily rose to more than 150 basis as an agency MBS. To do so, we first deduct the g-fee from points—a historical high that attracted much attention from the loan’s interest stream. We then compute the value of policymakers and commentators at the time. the remaining interest stream by interpolating MBS prices While the primary-secondary spread is a closely watched across coupons and subtracting the loan amount of $100.3 series, it is an imperfect proxy for the degree to which secondary- Chart 2 shows that the approximate net market value of a market movements are reflected in mortgage borrowing costs mortgage grew from less than 100 basis points (or $1 per (the “pass-through”) since, among other things, the secondary $100 loan) before 2009 to more than 350 basis points in yield is not directly observed, but model-determined, and thus the second half of 2012. Taken literally, the chart implies subject to model misspecification. Furthermore, mortgage that lender costs (other than the g-fee), lender profits, or a market pass-through depends on the evolution of the GSEs’ combination of the two must have increased by 300 basis guarantee fees (or “g-fees,” the price the GSEs charge for insuring points, or a factor of four, in five years. the loan) as well as on mortgage originators’ margins. To In this article, we first present a more detailed calculation understand changes in the extent of pass-through over time, it of originator profits and costs, and then attempt to explain is useful to track the two components separately. While g-fee their rise by considering a number of possible factors changes are easily observable, we argue that originator margins are best studied by tracking the different cash flows during and after the origination process, rather than by looking at the 3 For instance, assume that the mortgage note rate is 3.75 percent and primary-secondary spread (even after netting out g-fees). Indeed, the g-fee is 50 basis points, such that the remaining interest stream since originators are selling the loans, their margin depends on is 3.25 percent. Assuming that the 3.0 percent MBS trades at 102 and the 3.5 percent MBS trades at 104.5, the approximate market value of the price at which they can sell them, rather than the interest rate this mortgage in an MBS pool would then be simply the average of the on the security into which they sell the loans. two prices, 103.25, or 3.25 net of the loan principal.
18 The Rising Gap affecting them. In section 2, we begin with a general to close the loan.5 For example, for a borrower who wants discussion of the mortgage origination and securitization a $300,000, thirty-year fixed-rate mortgage, the originator process, and how originator profits are determined. may offer a 3.75 interest rate, known as the “note rate,” with Here, we include a detailed discussion of the valuation of the borrower paying $3,000 (or 1.0 percent) in closing costs. revenues from servicing and points as well as costs from If the borrower and originator agree on the terms, then the g-fees, based on standard industry methods. Next, in originator will typically guarantee these terms for a “lock-in section 3 we use these methods to derive a time series of period” of between thirty and ninety days, and the borrower average originator profits and unmeasured costs (OPUCs) will officially apply for the loan. for the period 1994-2012, which largely reflects the time- During the lock-in period, the originator processes the series pattern of Chart 2. We then compare OPUCs and loan application, performing such steps as verifying the the primary-secondary spread as measures of mortgage borrower’s income and the home appraisal. Based on the market pass-through. Finally, in section 4 we turn to results of this process, borrowers may ultimately not qualify possible explanations for the increase in OPUCs, including for the loan, or for the rate that the originator initially putback risk, changes in the valuation of mortgage servicing offered. In addition, borrowers have the option to turn rights, pipeline hedging costs, capacity constraints, market down the loan offer, for example, because another originator concentration, and streamline refinancing programs. While may have offered better loan terms. As a result, many loan some of the costs faced by originators may have risen over applications do not result in closed loans. These “fall-outs” the period 2008-12, we conclude that a large component of fluctuate over time and present a risk for originators, as we the rise in OPUCs remains unexplained by cost increases discuss in more detail in section 4. alone, suggesting that originators’ profits likely increased Originators have a variety of alternatives to fund loans: over this period. We then discuss possible sources of the they can securitize them in the private-label MBS market or rise in profitability. Capacity constraints likely played a in an agency MBS, sell them as whole loans, or keep them on significant role in enabling originator profits, especially their balance sheets. In the following discussion, we focus on during the early stages of refinancing waves. Pricing power loans that are “conforming” (meaning that they fulfill criteria coming from refinancing borrowers’ switching costs could based on loan amount and credit quality, so that they are eligi- have been another factor sustaining originator profits.4 ble for securitization by the GSEs), and assume securitization in an agency MBS, meaning that this option either dominates or is equally profitable to the originator’s alternatives.6,7
2. Measuring the Profitability of 5 Throughout this article, we use the terms “lender” or “originator” somewhat Mortgage Originations imprecisely, as they lump together different origination channels that in practice operate quite differently. Currently, the most popular origination channel is the “retail channel” (for example, large commercial banks that lend directly), which accounts for about 60 percent of loan originations, up from around 40 percent over the period 2000-06 (source: Inside Mortgage Finance). The alternative “wholesale” channel consists of brokers and “correspondent” 2.1 The Origination and Securitization lenders. Brokers have relationships with different lenders that fund their loans, and account for about 10 percent of originations. Correspondent Process lenders account for 30 percent of originations, and are typically small independent mortgage banks that have credit lines from and sell loans The mortgage origination process begins when a borrower (usually including servicing rights) to larger “aggregator” or “sponsor” banks. Our discussion in this section applies most directly to retail loans. seeks a quote for a loan, either to purchase a home or to 6 refinance an existing mortgage. Based on the borrower’s The fraction of mortgages that are not securitized into agency MBS has steadily decreased in recent years, according to Inside Mortgage Finance: credit score, stated income, loan amount, and expected while the estimated securitization rate for conforming loans ranged loan-to-value (LTV) ratio, an originator offers the borrower from 74 to 82 percent over the period 2003-06, it has varied between a combination of an interest rate and an estimate of the 87 and 98 percent since then (the 2011 value was 93 percent). The private- amount of money the borrower will need to provide up front label MBS market has effectively been shut down since mid-2007, with the exception of a few deals involving loans with amounts exceeding the agency conforming loan limits (“jumbo” loans). 7 Our discussion throughout this article applies directly to conventional mortgages securitized by the GSEs Fannie Mae and Freddie Mac; the process 4 Importantly, this article focuses on longer-term changes in the level of of originating Federal Housing Administration (FHA) loans and securitizing originator profits and costs, rather than on the high-frequency pass-through them through Ginnie Mae is similar, but with some differences (such as of changes in MBS valuations to the primary mortgage market. insurance premia) that we do not cover here.
FRBNY Economic Policy Review / December 2013 19 A key feature of an agency MBS is that principal and interest E 1 payments for these securities are guaranteed by the GSEs.8 The Example of a TBA Price Screen GSEs charge a monthly flow payment, the g-fee, which is a fixed fraction of the loan balance. Flow g-fees do not depend on loan characteristics but may differ across loan originators. Until 2012, flow g-fees averaged approximately 20 basis points per year, but during 2012 they rose to about 40 basis points, reflecting a Congressionally mandated 10-basis-point increase to fund the 2012 payroll tax reduction and another 10-basis-point increase mandated by the Federal Housing Finance Agency (FHFA). As we discuss below, originators can convert all or part of the flow g-fee into an up-front premium by “buying down” the g-fee. Alternatively, they can increase the flow g-fee and receive an up- front transfer from the GSE by “buying up” the g-fee. Since 2007, the GSEs have also been charging a separate up-front premium due upon delivery of the loan, known as the loan-level price adjustment (LLPA).9 The LLPA contains a fixed charge for all loans (currently 25 basis points) known as Source: Bloomberg L.P. an adverse-market delivery charge, as well as additional loan- Notes: Prices are quoted in ticks, which represent 1/32nd of a dollar; for instance, 103-01 means 103 plus 1/32 = $103.03125 per $100 par value. specific charges that depend on loan characteristics such as e “+” sign represents half a tick (or 1/64). Quotes to the le of the the term of the loan, the LTV, and the borrower’s FICO score. “/” are bids, while those to the right are asks (or oers). For instance, as of early 2013, the LLPA for a borrower with a FICO score of 730 and an LTV of 80 was 50 basis points (for a thirty-year fixed-rate loan; the charge is waived for loans with a term of fifteen or fewer years). Together with the 25-basis- and obligation to service the loan (which involves, for instance, point adverse-market delivery charge, this implies that the loan collecting payments from the borrower) and can be seized by the originator pays an up-front fee equal to 0.75 percent of the loan guaranteeing GSE if the servicer becomes insolvent. Servicing amount. Thus, the total up-front transfer between the originator income in excess of 25 basis points—10 basis points in this and GSE consists of the LLPA plus or minus potential g-fee example—is known as “excess servicing,” and is a pure interest buy-ups or buy-downs, which can be either positive or negative. flow. One might surmise here that a loan in a 3.0 percent pool For simplicity, our discussion assumes that the transfer from the must have a rate of 3.65 percent or higher (3.0 plus 40 basis originator to the GSE is positive and refers to it as an “up-front points for the g-fee plus 25 basis points for base servicing), insurance premium” (UIP). but recall from above that the originator can buy down the Once an originator chooses to securitize the loan in an g-fee so, in fact, the minimum note rate in a 3.0 percent pool agency MBS pool, it can select from different coupon rates, is 3.25 percent. In practice, for a mortgage of a given note rate, which typically vary by 50-basis-point increments. The note rate originators compare the profitability of pooling it in different on the mortgage, for example, 3.75 percent, is always higher than coupons, as described below. the coupon rate on an agency MBS, for example, 3.0 percent. Originators typically sell agency loans in the so-called TBA Who receives the residual 75-basis-point interest flow? (to-be-announced) market. The TBA market is a forward market Assuming the originator does not buy up or down the g-fee, in which investors trade promises to deliver agency MBS at approximately 40 basis points go to the GSEs (as of early 2013), fixed dates one, two, or three calendar months in the future. For leaving 35 basis points of “servicing income.” The GSEs require concreteness, Exhibit 1 displays TBA prices from Bloomberg at the servicer to collect at least 25 basis points in servicing income, 11:45 a.m. on January 30, 2013. At this time, investors will pay + ≈ known as “base servicing.” Base servicing is tied to the right 102 14 /32 102.45 for a 3.0 percent Fannie Mae (here denoted FNCL) MBS for April settlement. To understand the role of the TBA market, suppose that Bank A expects to have $100 million 8 If the loan is found to violate the representations and warranties made by the of 3.5 percent note rate mortgages available for delivery in seller to the GSEs, the GSEs may put the loan back to the seller. April. In order to hedge its interest rate risk, Bank A will then 9 LLPA is the official term used by Fannie Mae; Freddie Mac calls the sell $100 million par of 3.0 percent pools “forward” in the TBA corresponding premium “postsettlement delivery fee.” The respective fee grids can be found at www.fanniemae.com/content/pricing/llpa-matrix.pdf and market at a price of $102.45 per $100 par, to be delivered on www.freddiemac.com/singlefamily/pdf/ex19.pdf. the standard settlement day in April. Over the following weeks,
20 The Rising Gap E Mortgage Loan Securitized in an Agency MBS and Sold in TBA Market: The Money Trail
Government-Sponsored Borrower Originator Enterprise Investor
Cash ow from investor • Receives $100 for loan Origination Cash Flow: • Pays TBA(r ) to borrower Ω = TBA(r ) + coupon • Pays points to originator coupon • Receives UIP for loan (at time of for closing costs points – 100 – UIP origination)
Cash ow from borrower Servicing Cash Flow: • Receives rcoupon to investor • Pays rnote σ = • (during life of t Receives g-fee • Receives principal • Pays principal repayment r - g-fee - r loan; expressed in note coupon repayment annual terms)
OPUCs = Ω + PV(σ1, ...) 100 - points PV(rcoupon) = TBA(r ) - UIP Net bene t coupon UIP + PV(g-fee) + PV(principal repayment) - PV(rnote ) - 100 + points + - PV(principal repayment) - TBA(rcoupon) PV(rnote - g-fee - rcoupon )
Note: TBA(rcoupon) is the price of a mortgage-backed security (MBS) with coupon rate rcoupon in the “to-be-announced” market; UIP is up-front insurance premium (consisting of loan-level price adjustments plus or minus potential g-fee buy-ups or buy-downs); PV is present value.
Bank A assembles a pool of loans to be put in the security and From the investor’s payment, an originator funds the loan and delivers the loans to Fannie Mae, which then exchanges the pays any UIP to Fannie Mae.11 Together with points received loans for an MBS. This MBS is then delivered by Bank A on from the borrower, the cash flow to the originator when the the contractual settlement day to the investor who currently loan is made equals: owns the TBA forward contract in exchange for the promised Ω _ Origination cash flow (1) $102.45 million. A key feature of a TBA trade is that at the time of = TBA(r ) + points − 100 − UIP. trade, the seller does not specify which pools of loans it will deliver coupon to the buyer—this information is “announced” only shortly before the trade settles. As a consequence, market participants generally Through the life of the loan (middle panel of Exhibit 2), price TBA contracts under the assumption that sellers will deliver a borrower pays the note rate, rnote , from which Fannie Mae 10 the least valuable—or “cheapest-to-deliver”—pools at settlement. deducts the g-fee and the investor gets rcoupon, leaving servicing cash flow to the originator equal to:
_ = − − σt _ servicing cash flowt rnote g-fee rcoupon. (2) 2.2 How Does an Originator Make Money on the Transaction? Originator profits per loan are the sum of profits at origination (equation 1) and the present value (PV) of the A mortgage loan involves an initial cash flow at origination servicing cash flow (equation 2) less all marginal costs (other from investors to the borrower, and subsequent cash flows than the g-fee) of originating and servicing the loan, which we from the borrower to investors as the borrower repays the call “unmeasured costs.” Thus, loan principal and interest. Exhibit 2 maps these cash flows for a mortgage loan securitized in a Fannie Mae MBS and sold = + σ σ originator profits Ω PV( 1, 2,…) (3) in the TBA market. The top panel shows the origination cash − unmeasured costs. flow, which involves the investor paying priceTBA (rcoupon) to the originator in exchange for an MBS with coupon rate rcoupon. 11 Here and below, “originator” refers to all actors in the origination and servicing process, that is, if a loan is originated through a third-party 10 See Vickery and Wright (2013) for an overview of the TBA market. mortgage broker, for instance, the broker will earn part of the value.
FRBNY Economic Policy Review / December 2013 21 In our empirical exercise below, we study the sum of profits IO Strip Prices or Coupon Swaps and unmeasured costs, which is what we can observe: Servicing income can be thought of as an interest-only (IO) originator profits and (4) strip, which is a security that pays a flow of interest payments, = + but no principal payments, to investors as long as a loan is unmeasured costs (OPUCs) Ω PV(σ1, σ2,…). active.12 The main driver of the valuation of an IO strip is In later sections of the article, we attempt to assess to what the duration of the loan—an IO strip is far more valuable if extent changes in unmeasured costs can explain fluctuations one expects the borrower to prepay in five years as opposed in OPUCs. to one year; as in the latter case, interest payments accrue We next consider a specific transaction to illustrate how for a much shorter time period. One simple way to value the computations in Exhibit 2 are done in practice. Consider IO strips is to construct them from TBA securities through a loan of size $100 with a note rate of 3.75 percent locked in coupon swaps. For example, going long on a 3.5 percent on January 30 for sixty days by a borrower with a FICO score MBS and short on a 3.0 percent MBS generates interest cash of 730 and an LTV ratio of 80. The borrower agrees to pay flows of 50 basis points with prepayment properties that 1 point to the originator for the closing, and the originator correspond roughly to loans in 3.0 and 3.5 pools. According sells the loan into a TBA security with a 3.0 percent coupon to Exhibit 1, that 50-basis-point IO strip for April settlement for April settlement to allow sixty days for closing. Assuming would cost 2 11/32 (104 25+/32 minus 102 14+/32) ≈ 2.34. the loan closes, how high are the OPUCs? Since our originator has only 35 basis points of servicing, Computing the net revenue at origination, Ω, is relatively the coupon swap method would value servicing straightforward. According to Exhibit 1, investors pay rights at 35/50 × 2.34 ≈ 1.6, resulting in OPUCs of $102.45 for every $100 of principal in a TBA security with 2.7 + 1.6 = 4.3 points.13 This method ignores the fact that base servicing Computing the net revenue at generates other revenues, such as float income, in addition to the IO strip. To account for this additional value, it origination, Ω, is relatively is often assumed that the base servicing is worth more straightforward.… Valuing the stream than the present value of the IO strip. Assuming that base servicing is worth, for example, 25 percent more than of servicing income after origination, excess servicing would yield a PV of servicing income of + × ≈ (σ1 , σ2 , …), is more complicated. (25 × 1.25 10)/50 2.34 1.9, so that OPUCs would equal 2.7 + 1.9 = 4.6 points. Another shortcoming of the coupon swap method is that the a 3.0 percent coupon. As discussed earlier, the up-front coupon swap reflects differences in assumed loan characteristics insurance premium from the LLPA (and assuming no g-fee across coupons. For example, TBA prices may reflect the buy-up/-down) at the time was 0.75 percent of the loan fact that higher coupons are older securities having different (or 0.75 points). The originator collects 1 point from the prepayment characteristics. These differences will distort the borrower, remitting $100 for the loan, yielding Ω = 2.7 points. valuations of interest streams from the coupon swaps.14 Valuing the stream of servicing income after origination,
(σ1, σ2, …), is more complicated. For now, we assume that the originator does not buy up or down the g-fee—a decision that we will revisit below. This means that from the borrower’s Constant Servicing Multiples interest flow of 3.75 percent, the GSEs collect 40 basis points, An alternative method for valuing servicing flows is to use while the investors get 3.0 percent, leaving 35 basis points in fixed accounting multiples that reflect historical valuations of flow servicing income, σt , decomposed into 25 basis points of base servicing and 10 basis points of excess servicing. There are a number of alternative ways to determine the present 12 Another way to describe an IO strip is as an annuity with duration equal to value of these flow payments: the life of the loan. 13 This is the method implicitly used in the back-of-the-envelope calculation in Chart 2, except that there we ignored points paid by the borrower. 14 As an illustration, a 50-basis-point IO strip from a new 4.0 percent loan may not be worth as much as the price difference between the 3.5 and the 4.0 TBAs suggests, because the 4.0 TBAs may consist of loans that are older or credit impaired and thus prepay more slowly.
22 The Rising Gap servicing. In the industry, the base servicing multiple is often know whether the ones that trade are systematically more or assumed to be 5x, meaning that the present value of 25 basis less valuable than the ones that originators hold. points equals 1.25, while excess servicing is assumed to be valued at 4x, so that the value of the excess servicing in our example is 0.40. Using these servicing multiples, we see that the servicing income in our example is worth 1.65, meaning 2.3 Best Execution that OPUCs for this loan would be 2.7 + 1.65 = 4.35 points. Lenders can decide to securitize a loan into securities having different coupons, which involves different origination and servicing cash flows. The strategy that maximizes OPUCs is Buy-ups known in the industry as “best (or optimal) execution.”15 As mentioned above, originators can convert the g-fee into an Thus far, we have assumed that the originator securitizes up-front premium, or vice versa, using buy-ups and buy-downs. the loan in a 3.0 coupon. However, since the note rate is 3.75, 16 A buy-up means that the flow g-fee increases, but to compensate, the originator could alternatively sell it in a 3.5 coupon. the GSE will reduce the UIP (or, in case it is negative, transfer Given that the originator must retain 25-basis-point base money to the originator upon delivery of the loan). Thus, buying servicing, such a choice would require buying down the up the g-fee is a way to reduce the flow servicing income and entire 40-basis-point g-fee, meaning that instead of any flow increase income at the time of origination. payment to the GSE, the originator pays the full insurance The GSEs offer a buy-up multiple, which is communicated premium up front. Exactly like the buy-up multiple discussed to originators (but not otherwise publicly known), and varies above, the GSEs also offer a (higher) buy-down multiple, over time, presumably with the level of the coupon swap. If, which determines the cost of this up-front payment. for example, the buy-up multiple is 3x, then a 10-basis-point Using the prices in Exhibit 1, we note that the price increase in the g-fee reduces UIP by 30 basis points, lowering of a 3.5 TBA coupon is 104 24+/32=104.77, meaning σ Ω that changing coupons would increase loan sale revenues t by 0.1 and raising by 0.3. Note that only excess servicing, σ by 2.32 points. If we assume the buy-down multiple t, -0.25, can be “monetized” this way, while 25 basis points of base servicing still need to be retained and valued by the equals 7, then UIP would increase by 2.8 points relative to originator. If we assume a base servicing multiple of 5x, as the 3.0 coupon case. Ω is thus equal to 2.22, or 0.48 less than above, then buying up the g-fee by 10 basis points would lead it would be for the 3.0 coupon case. Meanwhile, servicing σ = to OPUCs of 3.0 + 1.25 = 4.25. income is now simply t 0.25, as the flow g-fee has been The buy-up multiple provides a lower bound on the bought down to zero, and with an assumed base servicing valuation of excess servicing—the originator (or some other multiple of 5x, OPUCs for this execution would equal servicer) may value it at a higher multiple; but if it does not, 2.22 + 1.25 = 3.47. it can sell its excess servicing to the GSEs. To what extent Comparing this OPUC value with the “constant originators want to take advantage of this option depends on servicing multiples” case above, we see that pooling into a number of factors. For example, as we discuss in section 4.1, the 3.0 coupon would generate higher OPUCs than the the upcoming implementation of Basel III rules may require 3.5 coupon and thus would be best execution for a mortgage banks to hold additional capital against mortgage servicing with the 3.75 percent note rate. assets, which may lower their effective valuation of servicing However, this conclusion is sensitive to a number of income. By buying up the g-fee, these banks can turn servicing assumptions—in particular, the valuation of excess servicing 17 cash flows that are subject to additional regulatory capital and the buy-down multiple. As shown in Table 1, pooling in charges into cash. Another potential factor is the originator’s the higher coupon becomes more attractive as the buy-down beliefs about the prepayment properties of a pool of loans. For multiple decreases or the excess servicing multiple decreases. example, if a lender believes that the expected lifetime of a pool is shorter than average, it may choose to buy up the g-fee. 15 See Bhattacharya, Berliner, and Fabozzi (2008) for an extensive discussion of pooling economics and mortgage pricing that also includes nonagency Market Prices of Servicing Rights securitizations. 16 The originator could also place the loan in a 2.5 percent or lower coupon— Finally, there is an active market for trading servicing rights, the only restriction is that the note rate cannot be more than 250 basis points which can be sold by originators at origination or well after- above the coupon. ward. One can use market prices to value servicing rights, but 17 As base servicing always needs to be retained, its valuation does not affect since not all servicing rights change hands, it is difficult to best execution—it shifts OPUCs up or down equally for all coupons.
FRBNY Economic Policy Review / December 2013 23 Table 1 Table 2 Dependence of Best Execution on Excess Servicing Example of a Mortgage Rate Sheet and Buy-Down Multiples
Lock-in Period Excess Servicing Buy-Down OPUCs(3.0) OPUCs(3.5) Note Rate Fifteen Days Thirty Days Sixty Days Multiple Multiple (Points) 4.750 (3.956) (3.831) (3.706) 4x 7x 4.35 3.47 4.625 (3.831) (3.706) (3.581) 4x 5x 4.35 4.27 4.500 (3.706) (3.581) (3.456) 3x 5x 4.25 4.27 4.375 (3.331) (3.206) (3.081) Sources: Bloomberg L.P.; authors’ calculations. 4.250 (3.081) (2.956) (2.831) Note: OPUCs are originator profits and unmeasured costs. 4.125 (1.831) (1.706) (1.581) 4.000 (1.456) (1.331) (1.206) 3.875 (1.081) (0.956) (0.831) 3.750 (0.831) (0.706) (0.581) 2.4 Rate Sheets and Borrower Choice 3.625 (0.081) 0.044 0.169 3.500 0.794 0.919 1.044 Until now, we have taken the borrower choice as given—the 3.375 1.669 1.794 1.919 borrower pays 1 point at origination and is offered a note rate of 3.250 2.544 2.669 2.794 3.75. However, from our OPUC calculations, it is clear that there 3.125 3.919 4.044 4.169 are other combinations of note rate and points that would be Source: www.53.com/wholesale-mortgage/wholesale-rate-sheets.html on equally profitable for the originator. For example, if the borrower January 30, 2013. paid a note rate of 4.0 instead, and the originator still pooled Notes: Figures are in percentage points of the loan amount. Loan type is a the loan into a 3.0 coupon, then excess servicing would increase thirty-year fixed-rate loan. Column 1 shows the annual interest rate to be by 25 basis points, leading to 1 point higher revenue under paid by the borrower over the life of the loan. Columns 2-4 show the points an excess servicing multiple of 4x. Therefore, the originator the borrower needs to pay up front to obtain the interest rate in column 1, for different lock-in periods. Parentheses denote negative figures. could maintain its profit margin by offering the borrower a combination of 0 points at closing and a note rate of 4.0.18 Indeed, originators offer borrowers precisely these sorts of alternatives between closing costs and rates. Table 2 shows throughout the example. If the borrower wanted a lower note part of a rate sheet provided by a bank to a loan officer on rate, for example, 3.5 percent, then the closing costs would January 30, 2013.19 The entries in the table are “discount rise by 1.044 − (-0.581) = 1.625, or from 1 to 2.625 points. points,” which are points paid by the borrower at closing to Alternatively, by choosing a rate of 4.125 percent, the lower the note rate on the loan. For example, assume that the borrower could get a rebate of 1.581 points and would pay total closing fees the originator would charge the borrower nothing at closing. without any discount points would equal 1.58 points— As shown in the rate sheet, there is no single “mortgage sometimes referred to as “origination points.” These fees rate.” Rather, a large number of different note rates are include application processing costs, compensation for the available to borrowers on any given day, typically in loan officer, and also the LLPA (0.75 points in our example), increments of 0.125.20 Originators simply change the number which is usually charged directly to the borrower. of discount points offered for the different note rates one or Our baseline borrower has a sixty-day lock-in period and more times a day, reflecting secondary-market valuations a note rate of 3.75 percent; accordingly, based on the rate (TBA prices), servicing valuations, and GSE buy-up/ sheet, the borrower is contributing -0.581 discount points. buy-down multiples.21 This means that the bank is actually paying the borrower cash up front (often referred to as a “rebate”), which reduces 20 That said, banks will often quote a headline mortgage rate, which is closing costs from 1.58 points to the 1 point assumed generally the lowest rate such that the number of discount points required from the borrower is “reasonable” (this rate is sometimes referred to as the “best-execution” rate for the borrower, not to be confused with the 18 In fact, the 4.0 note rate might increase the profit margin, because it would originator’s best execution). In the example rate sheet, this rate would likely be potentially alter the best-execution coupon. 3.75 or 3.625, as going below 3.625 requires significant additional points from the borrower. 19 Actual sample rate sheets can be found, for instance, at www.53.com/ wholesale-mortgage/wholesale-rate-sheets.html. Most lenders do not make 21 The set of available note rates on a given day generally depends on which their rate sheets available to the public. MBS coupons are actively traded in the secondary market.
24 The Rising Gap 2.5 Summary: Trade-offs, Trade-offs Chart 3 Average Effective Guarantee Fee Everywhere Basis points 55 As shown in the preceding discussion, the different actors in the origination and securitization process have a number of 50 trade-offs available to them. Borrowers can decide between 45 paying more points up front and paying a higher interest rate 40 later. Originators can choose between different coupons into which to pool a loan, which imply different origination and 35 servicing cash flows; in addition, as part of this decision, origi- 30 nators can choose whether to pay the GSE insurance premium 25 up front or as a flow. Finally, investors can choose to invest 20 in securities with different coupons, with higher coupons requiring a larger initial outlay, but subsequently generating 15 2002 04 06 08 10 12 higher flow payments. Investor demand for different coupons, which reflects their prepayment and interest rate projections, Source: Fannie Mae SEC Forms 10-K and 10-Q, various issues through 2012:Q4. ultimately affects originators’ best-execution strategies and thus the point-rate grid offered to borrowers.
To circumvent this issue, and also for the sake of 3. Measuring OPUCs over Time simplicity, our baseline calculations use fixed multiples of 5x for base servicing, 4x for excess servicing, and 7x for Our goal in this section is to derive an empirical measure buy-downs.22 These are commonly assumed values in of average OPUCs (equation 4) for thirty-year fixed-rate industry publications. Later, we explore the sensitivity of mortgages for the period 1994 to 2012. To do so, we need to OPUCs to alternative assumptions. make a number of assumptions. Finally, we do a best-execution calculation, considering First, rather than valuing each possible loan note rate, we three differentTBA coupons (using back-month prices) value a hypothetical mortgage having a note rate equal to the into which the mortgage could potentially be pooled.23 The survey rate from Freddie Mac’s Primary Mortgage Market highest coupon is set such that it requires the originator to Survey, at weekly frequency. We also use the weekly time buy down some or all of the g-fee up front, while instead, series of average points paid from the same survey. for the other two possible coupon options, the originator Second, rather than accounting sepa- retains positive excess servicing because the loan’s interest rately for LLPAs and the flow g-fee, we use an “effective” g-fee, payment is more than sufficient to cover theg-fee and base which assumes that LLPAs are paid over the life of the loan, servicing.24 The best execution among the three options as reported in Fannie Mae’s Securities and Exchange Com- determines our OPUC value for the week in question. mission Form10-Q filings. The average size of the effective Before turning to the weekly OPUC time series, we report g-fee is shown in Chart 3. In our calculations, we incorporate in Table 3 a detailed OPUC calculation on a given day. We anticipated changes in g-fees. In particular, the 10-basis-point can infer, from the bottom of the table, that the mid-coupon increases that came into effect on April 1, 2012, and Decem- execution is optimal in this example. ber 1, 2012, are assumed in our calculations to apply to loans originated January 1 and September 1, respectively, which is right after the increases were announced. Third, as explained above, we need to value the servicing 22 We assume the buy-up multiple to be smaller than 4x, such that, in our income flow. The coupon swap method has the advantage of calculations, buy-ups are never used. being based on current market prices that reflect changes in 23 The use of back- rather than front-month TBA price contracts reflects the the duration of the cash flows. But, as mentioned earlier, the originators’ desire to hedge price movements during the lock-in period, as discussed in more detail in section 4. coupon swap may also reflect differences in assumed loan 24 characteristics across coupons; therefore, it may be a poor Depending on the mortgage rate, pooling into the highest candidate coupon may not actually be a possibility—as explained, the mortgage rate proxy for the value of an interest strip from a new loan. needs to exceed the coupon rate by at least 25 basis points.
FRBNY Economic Policy Review / December 2013 25 Table 3 Example of OPUCs Best-Execution Calculation
TBA Coupon (Percent) 3.5 4.0 4.5 (1)
Coupon-independent inputs (percent) Mortgage rate 4.78 4.78 4.78 (2) Points 0.7 0.7 0.7 (3) Effective g-fee 0.261 0.261 0.261 (4) Base servicing 0.25 0.25 0.25 (5)
Excess servicing 0.769 0.269 -0.231 (6) = (2) − (1) − (4) − (5)
Coupon-specific inputs (dollars per par value) TBA price (back-month) 97.55 99.95 101.67 (7) Value of base servicing 1.25 1.25 1.25 (10) = 5 × (5) Value of excess servicing 3.08 1.08 (11) = 4 × (6) if (6) > 0 G-fee buy-down -1.62 (12) = 7 × (6) if (6) < 0
Revenues from TBA sale less payout to borrower -1.75 0.65 2.37 (13) = (7) − (100 − (3)) Value of servicing net of g-fee 4.33 2.33 -0.37 (14) = (10) + (11) + (12)
OPUCs By coupon 2.58 2.98 2.00 (15) = (13) + (14) Best-execution 2.98 (16) = max(15) if (2) − (1) > .25
Source: Authors’ calculations. Note: Calculation is for April 30, 2009. OPUCs are originator profits and unmeasured costs; TBA is “to-be-announced.”
Chart 4 3.1 Results Originator Profits and Unmeasured Costs, 1994-2012 The weekly OPUC series over the period 1994 to 2012 is Dollars per $100 loan shown in Chart 4. The series averaged about $1.50 between 6 1994 and 2001, then temporarily increased to the $2.00-$3.00 range over 2002-03, before declining again 5 and remaining below $2.00 for most of the period 2005-08. Weekly 4 The OPUC measure jumped dramatically to more than $3.50 in early 2009 and then again in mid-2010. Most notably, Eight-week 3 rolling window however, it increased further over 2012, and reached highs of more than $5 per $100 loan in the second half of the year, 2 before declining again toward the end of 2012. As shown in the back-of-the-envelope calculation in 1 Chart 2, the higher valuation of loans in the MBS market is 0 the main driver of the increase in OPUCs toward the end 1994 96 98 00 02 04 06 08 10 12 of our sample period. Relative to that figure, the increase in OPUCs over 2009-12 in Chart 4 is less dramatic; this is Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; authors’ calculations. because the earlier calculation implicitly valued servicing through coupon swaps, which were very low in early 2009 but relatively high since 2010. In contrast, in Chart 4 we have
26 The Rising Gap 25 assumed constant multiples. As we discuss in more detail Table 4 below, servicing right valuations appear to have declined, Issuance of Various GSE Thirty-Year Fixed-Rate Pool rather than increased, over the past few years, supporting the Types, January–October 2012 use of fixed multiples rather than coupon swaps. When interpreting the OPUC series, it is important to keep Balance Loan Balance Count Pool Type (Millions of Dollars) Count (Percent) (Percent) in mind a few notes. First, the measure uses data on thirty-year conventional fixed-rate mortgage loans only and therefore TBA 379,763 1,347,516 59 46 a bears no direct information on other common types of loans, MHA 124,779 559,180 20 19 such as fifteen-year fixed-rate mortgages, adjustable-rate Loan balanceb 97,161 867,628 15 30 mortgages, Federal Housing Administration loans, or jumbos. Other specifiedc 36,588 138,735 6 5 Second, since the measure uses survey rates/points and Total 638,292 2,913,059 100 100 average g-fees, our OPUC series is an average industry Sources: Fannie Mae; Freddie Mac; 1010data; Amherst Securities. measure rather than an originator-specific one. In addition, Note: GSE is government-sponsored enterprise. TBA is “to-be- rates and points may be subject to measurement error that announced.” MHA is the Making Home Affordable program. could distort the OPUC measure at high frequency, although aIncludes pools that are 100 percent refi with 80
FRBNY Economic Policy Review / December 2013 27 C points.28 Importantly, the impact of potential prepayment Price of Lowest Fannie Mae TBA Thirty-Year model misspecification on yields is amplified when the Coupon with Sizable Issuance security trades significantly above (or below) par because
Monthly average price (in dollars) the yield on the security depends on the timing of the 106 amortization of the bond premium. 104 A better way to think about pass-through is to look directly at what happens with the money paid by an 102 investor in the secondary market—does it go to borrowers, 100 originators, or the GSEs (either up front, or through equivalent flow payments)? The purpose of the OPUC 98 measure is to track how many dollars (per $100 loan) get 96 absorbed by originators, either to cover costs other than the g-fee, or as originator profits.29 G-fees also contribute 94 to the overall cost of mortgage credit intermediation— 92 increasing these fees means that less money goes to 2000 02 04 06 08 10 12 borrowers (or equivalently, that they need to pay a Source: eMBS; JPMorgan Chase. higher rate). So, full pass-through of secondary-market movements to borrowers would require OPUCs and g-fees Notes: TBA is “to-be-announced.” “Sizable issuance” means that the coupon accounts for at least 10 percent of total issuance in to remain constant (or, alternatively, a rise in g-fees would that month. need to be offset by a decrease in OPUCs). In panel A of Chart 6, we conduct a counterfactual exercise in which we compute a hypothetical survey note rate during 2012, assuming that either the OPUCs only First, the yield on any MBS is not directly observable, (dark blue line), or both the OPUCs and the g-fee (light because the timing of cash flows depends on prepayments. blue line), had stayed at their average levels in 2011:Q4.30 Therefore, the calculation of the yield is based on the MBS The comparison of the light blue line with the black line, price and cash flow projections from a prepayment model, the actual realized mortgage rate, shows that had the cost which itself uses as inputs projections of conditioning of mortgage intermediation stayed constant relative to variables (for example, interest rates and house prices). In 2011:Q4, mortgage rates during 2012 would at times have addition, for TBA contracts, the projected cash flows and been substantially lower, with a maximum gap between the the yield also depend on the characteristics of the assumed two rates of 55 basis points in early October 2012. cheapest-to-deliver pool. The resulting yield is thus subject Comparing the black line with the dark blue line (holding to errors due to model misspecification. only OPUCs fixed but letting g-fees increase), we note that Second, the primary-secondary spread typically relies on over most of 2012, much of the gap between the actual the theoretical construct of a “current coupon MBS.” The and counterfactual rate derives from the rise in OPUCs. current coupon is a hypothetical TBA security that trades at par and has a yield meant to be representative of those 28 Additionally, the current coupon is typically based on front-month contract on newly issued securities.27 Historically, this par contract prices, while a more accurate measure would use back-month contracts, because loans that rate-lock today are typically packaged into TBAs at least has usually fallen between two other actively traded TBA two months forward. coupons; however, in recent times, even the lowest coupon 29 It is important to keep in mind that changes in the secondary yield, even with nontrivial issuance has generally traded significantly if correctly measured, do not necessarily translate one-to-one into changes above par (Chart 5). As a result, the current coupon rate in originator margins, which are determined by the TBA prices of different is obtained as an extrapolation from market prices, rather coupons (which in turn determine optimal execution), and also by points paid by the borrower. The primary-secondary spread, even net of g-fees, is than a less error-prone interpolation between two traded thus at best an imprecise measure of originator margins and profitability. 30 The effective g-fee in our calculation for 2011:Q4 is 28.8 basis points, which then increases to 38.9 basis points for the period January-March 2012 (as the announced increase effective April 1, 2012, is assumed to already be 27 An alternative is to calculate the yield on a particular security, which relevant for loans originated at that point), 40.3 basis points for the period may trade at a pay-up to the cheapest-to-deliver security. However, such a April-June 2012, 41.8 basis points for July and August, and then increases calculation is still subject to other model misspecification and would not be by another 10 basis points, to 51.8 basis points, for the rest of 2012 as the representative of the broad array of newly issued securities. December 1 g-fee increase becomes relevant to pricing.
28 The Rising Gap C such as the average OPUCs over all of 2011, the dark blue and Counterfactual Paths of Mortgage Rates over 2012 Percent light blue lines would be significantly lower. In panel B of Chart 6, we conduct a similar counterfactual Panel A: Holding OPUCs and g-fees constant at 2011:Q4 averages rate analysis, but using the primary-secondary spread as the 4.2 measure of the cost of mortgage intermediation. Holding Fixed-rate mortgage (FRM) 4.0 rate with constant OPUCs this spread (measured as the Freddie Mac survey rate minus 3.8
3.6 Actual FRM rate with FRM rate Over most of 2012, much of the gap 3.4 constant OPUCs and constant g-fee between the actual and counterfactual 3.2 rate derives from the rise in OPUCs. 3.0
2.8 2.6 the Bloomberg current coupon yield) constant, we again get a hypothetical mortgage rate under full pass-through. As Panel B: Holding primary-secondary spread constant at 2011:Q4 average shown in panel B, while the overall pattern is similar to the 4.2 counterfactual rate with constant OPUCs and g-fees in panel A, 4.0 the series in panel B is more volatile, with the gap between the counterfactual and actual rate spiking at 75 basis points in late 3.8 Actual September 2012. This volatility of the counterfactual rate and FRM rate 3.6 the presence of such large spikes illustrate the imperfect nature 3.4 of the primary-secondary spread as a pass-through measure.
3.2
3.0 FRM rate with constant primary-secondary spread 2.8 4. Potential Explanations for the 2.6 Rise in Costs or Profits Q1 Q2 Q3 Q4 The rest of the article explores in more detail factors that may Sources: Bloomberg L.P.; Freddie Mac; authors’ calculations. have driven the observed increase in OPUCs over the period Note: OPUCs are originator pro ts and unmeasured costs. 2008-12. On the cost side, we focus on changes in pipeline hedging costs, putback risk, and possible declines in the valuation of mortgage servicing rights. We also briefly discuss changes in loan production expenses. On the profit side, we Additionally, it is apparent that in times when rates are stable focus on potential increases in originators’ pricing power due or increasing, the counterfactual rate with constant OPUCs to capacity constraints, industry concentration, or switching tends to be close to the actual rate, and most of the gap costs for refinancers. between the black and the light blue lines comes from the higher g-fees (this is the case, for instance, toward the end of the year). It is during times when rates fall (secondary- market prices increase) that actual rates do not fall as much 4.1 Costs as they would with constant OPUCs. As we discuss later, this is consistent with originators having limited capacity, which means they can keep rates relatively high and make extra profits. That said, one should not necessarily interpret the Loan Putbacks counterfactual rate series as indicating “where rates should have been,” as this would require a judgment regarding Originators pay g-fees to the GSEs as an insurance premium; the “right” level of OPUCs. Here, we took the average over in exchange, the GSEs pay the principal and interest of the 2011:Q4 as our baseline, but if instead we took a lower value, loan in full to investors when the borrower is delinquent.
FRBNY Economic Policy Review / December 2013 29 However, mortgage originators or servicers are obligated to To account for the tighter underwriting standards on new repurchase nonperforming or defaulted loans under certain loans, we focus on the performance of GSE-securitized loans conditions, for example, when the GSEs establish that the loan from the 2005-08 vintages with origination FICO of at least did not meet their original underwriting or eligibility require- 720 and full documentation. Among those, “only” about ments, that is, if the loan representations and warranties are 8.8 percent have become sixty-or-more days delinquent, and flawed.31 The repurchase requests have increased rapidly since 5.5 percent have ended in foreclosure. Thus, because of today’s the 2008 financial crisis and have been the source of disputes more stringent underwriting guidelines for agency loans, our between originators and GSEs. The increased risk to origina- expectation in a stress scenario would be for delinquencies, tors that the loan may ultimately be put back to them has been and hence potential putbacks, to be roughly half as large, cited as a source of higher costs and thus OPUCs. relative to those experienced by the 2005-08 vintages. Further- How can we assess the magnitude of the contribution of more, we would expect the frequency of putback attempts to putback costs to OPUCs? To do so, one needs to imagine be roughly half as large for loans with full documentation as a stress scenario—not a modal one—with a corresponding for the overall population of delinquent loans. We obtain an estimate of the fraction of loans that the GSEs could attempt to force the lender to repurchase from The increased risk to originators that the Fannie Mae’s 2012:Q3 Form 10-Q, which states (on page 72) loan may ultimately be put back to them that as of 2012:Q3, about 3 percent of loans from the 2005-08 vintages have been subject to repurchase requests (compared has been cited as a source of higher with only 0.25 percent of loans originated after 2008). Thus, costs and thus OPUCs. given that repurchase requests are issued primarily conditional on a delinquency, we would anticipate repurchase requests in a stress scenario to be about one-quarter (0.5 delinquency default rate, and then assume fractions of putback attempts rate × 0.5 putback rate) as high as those recorded on the by the GSEs, putback success, and loss-given-defaults for 2005-08 vintage, or about 0.75 percent.34 servicers/lenders forced to repurchase the delinquent loan. Based on repurchase disclosure data collected from the To construct a ballpark estimate of the possible putback GSEs,35 it appears that about 50 percent of requests ultimately cost on new loans, we start from the experience of agency lead to buybacks of the loan. Furthermore, if we assume a loans originated during the period 2005-08. Based on a 50 percent loss-given-default (which seems on the high side), random 20 percent sample of conventional first-lien fixed-rate this would generate an expected loss to the lender/servicer of: loans originated during that period in the servicing data set of LPS Applied Analytics, we find that about 16.5 percent of 0.75 percent × 0.5 × 0.5 = 19 basis points GSE-securitized mortgages (value-weighted) have become sixty-or-more days delinquent at least once, and 11.5 percent This estimate, which we think of as being conservative 32 of them have ended in foreclosure. Importantly, these (given the unlikely repetition at this point of large house vintages include a substantial population of borrowers with price declines experienced by the 2005-08 vintages), would relatively low FICO scores, undocumented income or assets, imply a putback cost of 19 cents per $100 loan. This cost is or a combination of these factors. For instance, the median modest relative to the widening in OPUCs experienced over FICO score was around 735, while the 25th percentile the period 2008-12.36 That said, perhaps the “true” cost of was at 690. In 2012, however, the corresponding values on putback risk comes from originators trying to avoid putbacks 33 non-HARP loans were around 770 and 735, respectively. in the first place by spending significantly more resources on underwriting new loans or on defending against putback
31 It is also possible that originators need to repurchase incorrectly underwritten loans prior to a loan becoming delinquent. However, the repurchase of nondelinquent loans is likely less costly to originators. The 34 Without the assumption that full-documentation loans are less likely to rest of this section therefore focuses on repurchases of delinquent loans. be put back, the expected putback rate would be 1.5 percent, resulting in an expected loss of 37.5 basis points. 32 These statistics are as of November 2012. 35 Source: Inside Mortgage Finance. 33 Origination LTVs have not changed as dramatically: in 2012, approximately 16 percent of non-HARP loans had an LTV at origination above 80; this is only 36 Furthermore, the FHFA introduced a new representation and warrant slightly lower than during the period 2005-08. However, the fraction of loans framework for loans delivered to the GSEs after January 2013 that relieves with second liens was likely higher during the boom period. Also, in 2012 there lenders of repurchase exposure under certain conditions (for example, if the are no non-HARP Freddie Mac loans with incomplete documentation (this is loan was current for three years). This policy change should further reduce not disclosed in the Fannie Mae data, but is likely similar). the expected putback cost going forward.
30 The Rising Gap claims. Furthermore, the remaining risk on older vintages is Chart 7 larger than on new loans, and many active lenders are also Sensitivity of OPUCs to Alternative Assumptions still subject to lawsuits on nonagency loans made during the about Mortgage Servicing Right Multiples boom. It is unclear, however, why these claims on vintage Dollars per $100 loan loans should affect the cost of new originations. 5
4 Baseline Mortgage Servicing Rights Values multiples
The baseline OPUC calculation assumes constant servicing 3 multiples throughout the sample of 5x for base servicing and 4x for excess servicing flows. While these are commonly 1/2 multiples assumed levels, according to market reports, mortgage 2 servicing right (MSR) valuations have declined over the past MIAC multiples few years. In this section, we study the sensitivity of OPUCs 1 to alternative multiple assumptions. 2008 2009 2010 2011 2012 We obtain a time series of normal (or base) servicing Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; MIAC; authors’ multiples for production agency MBS coupons from the calculations. 37 company Mortgage Industry Advisory Corporation (MIAC). Notes: The data reflect an eight-week rolling window. MIAC is the These multiples declined from about x5 in early 2008 to about Mortgage Industry Advisory Corporation. 3.25x in November 2012.38 To evaluate the impact on OPUCs, we repeat our earlier calculation using the MIAC base multi- ples.39 The results are shown in Chart 7. Comparing the black (baseline) and dark blue (MIAC) lines, we see that the lower these rules makes it unclear how much the expected tighter multiple values reduce OPUCs by about sixty cents at the end regulatory treatment is already affecting MSR multiples. of 2012, a somewhat significant impact. Nonetheless, in order to assess an upper-bound impact Some commentators have attributed the decline in on OPUCs, we consider here a more stressed scenario multiples to a new regulatory treatment of MSRs under the than implied by the MIAC multiples. In this scenario, our 2010 Basel III accord. While the three U.S. federal banking baseline multiples are halved starting (for simplicity) with regulatory agencies released notices of proposed rulemaking the disclosure by the Basel Committee of the capital rules in to implement the accord on June 12, 2012, the introduction July 2010.41 The resulting eight-week-rolling OPUC series is of the new rules, originally set for January 2013, has been also depicted in Chart 7. As shown in the chart, following a postponed. Under the June 2012 proposal, concentrated halving of the MSR multiples, the implied OPUC declines are MSR investment will be penalized and will generally receive significant, but still not sufficient to explain the historically 40 a higher risk weighting. The long phase-in period for high OPUC levels in 2012. We conclude that lower multiples, while having a sizable 37 These multiples come from MIAC’s “Generic Servicing Assets” portfolio impact on OPUCs, can only partially offset their increase and are based on transaction values of brokered bulk MSR deals, surveys of over the past few years. market participants, and a pricing model. 38 Key drivers of servicing right valuations are expected mortgage prepayments—lower interest rates mean a higher likelihood that the servicing flow will stop due to an early principal payment—and, in the case of base servicing, varying operating costs in servicing the loan, for example, when loans become delinquent. Another important component is the magnitude of the float interest income earned, for instance, on escrow accounts. 39 We assume a 20 percent discount for excess servicing and keep the g-fee buy-down multiple unchanged at 7x. Also, as our MIAC series ends in November 2012, we assume that the multiple in December is identical to that in November. 40 MSRs will be computed toward Tier 1 equity only up to 10 percent of their value, and risk-weighted at 250 percent, with the rest being deducted from 41 In this alternative scenario, base servicing is now valued at 2.5x, while Tier 1 equity. This treatment is significantly more stringent than the status excess servicing is valued at 2x. (The GSE buy-down multiple is assumed to quo that risk-weights the MSRs at 100 percent and limits MSRs to 50 percent stay at 7x.) The optimal execution in this exercise again takes into account the of Tier 1 capital of banks (100 percent for savings and loans). lower levels of the multiples.
FRBNY Economic Policy Review / December 2013 31 Pipeline Hedging Costs Chart 8 Swaption Price Premia For loans that are securitized in MBS, the “mortgage pipeline” is the channel through which an originator’s loan commit- Basis points 225 ment, or rate-lock, is ultimately delivered into a security or terminated with a denial or withdrawal of the application. The 200 175 originators’ commitment starts with a rate-lock that typically Three-month, ranges between thirty and ninety days. This time window 150 five-year appears to have increased significantly in recent years. For 125 example, the time from application to funding for refinancing 100 applications increased from about thirty days in late 2008 75 to more than fifty days in late 2012 (as shown graphically in 50 section 4.2 below). One-month, Originators face two sources of risk while the loan is in 25 five-year the pipeline: changes in the prospective value of the loan due 0 2003 04 06 08 10 12 to interest rate fluctuations and movements in the fraction of rate-locks that do not ultimately lead to loan originations, Source: JPMorgan Chase. referred to as “fallouts.” The first risk—potential changes in the value of the loan due to interest rate movements—can be hedged by selling TBA contracts: at the time of the loan commitment, origina- offer) or voluntarily. Except for changes in lending standards and tors who are long a mortgage loan at the time of the rate-lock house prices, fluctuations in involuntary terminations are largely driven by idiosyncratic factors that are diversified for originators with large-enough portfolios. Movements in voluntary Originators face two sources of risk while terminations, on the other hand, are mostly due to primary rate the loan is in the pipeline: changes in dynamics: following the initial rate-lock, mortgage rates may fall, prompting borrowers to pursue a lower rate loan with either the prospective value of the loan due to the same or a different lender. Common ways to hedge this risk interest rate fluctuations and movements are to dynamically delta-hedge the position using TBAs, using in the fraction of rate-locks that do not mortgage options or swap options, or a combination of these (or other) strategies.42 To illustrate, we now consider a hedging ultimately lead to loan originations, example using at-the-money swaptions to gauge the magnitude referred to as “fallouts.” and time-series pattern of the interest rate hedging cost. Based on market reports and data from the Mortgage Bankers Association (MBA), normal fallout rates average can offset the position by selling the yet-to-be-originated about 30 percent, and we assume that an originator hedges loan forward in the TBA market. The calculation in section 3 as much using swaptions. Chart 8 shows the price premium already takes into account these hedging costs: when comput- in basis points for swaptions on a five-year swap rate with ing the OPUC measure, we use the back-month TBA contract expirations of one and three months. Conditional on a price that settles on average about forty-five days following 30 percent hedging strategy, the cost of protection, when the transaction. To the extent that originators may have been using a three-month expiration, would be about 0.3 x 40 basis able to sell into the front-month TBA market when the length points = 12 basis points, or a 12 cent impact on OPUCs. The of the pipeline was shorter, our calculations may understate extension in the length of the pipeline, which may have led OPUCs for earlier years by the price difference, or “drop,” originators to go from one-month to three-month expiration, between the two contract prices. Yet, this drop is typically also had a rather small impact on OPUCs. only about 20 basis points in price space. We conclude that the lengthening of the pipeline does not appear to have had a 42 Correspondent lenders, or small lenders that sell whole loans to the GSEs, significant economic impact on the cost of price hedging, and can manage the fallout risk by entering into “best-effort” locks with the buyer thus the rise in OPUCs experienced over the period 2008-12. of the loan. Under this arrangement, the originator does not need to pay a fine The second risk is due to movements in the fallout rate. for not delivering a mortgage that does not close, unlike under “mandatory delivery.” To compensate, the price offered by the buyer of the loan is lower. As discussed in section 2, borrowers’ terminations may occur Thus, in a sense, “best-effort” commitments allow (small) originators to involuntarily (if they do not ultimately qualify for the loan or rate “outsource” the hedging of fallout risk.
32 The Rising Gap More generally and beyond our specific example, implied C volatility and option price premia have declined significantly Originator Prots and Unmeasured Costs (OPUCs) since the fall of 2008, reflecting the lower rate volatility and MBA Application Index environment. While we do not explicitly consider other, more Dollars per $100 loan Index level complex hedging strategies, the lower volatility environment has 5 2000 MBA market volume likely also lowered the cost of these strategies. This is in contrast index (all applications) with the rise in OPUCs over this period. In sum, changing Right scale 4 1500 hedging costs does not appear to account for a significant portion of the rise in OPUCs, and at least the cost of hedging fallout risk may in fact have declined during the period 2009-12. 3 1000
2 500 Other Loan Production Expenses OPUCs Left scale A final possible cost-side explanation for the increase in 1 0 OPUCs is that other loan production expenses, including 1994 96 98 00 02 04 06 08 10 12 costs related to the underwriting of loans and to finding Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; Mortgage borrowers (sales commissions, advertising, and so on) have Bankers Association (MBA); authors’ calculations. increased substantially over the past few years. While it Note: e lines re ect eight-week rolling window averages. is difficult to obtain a variable loan cost series that can be easily mapped into the OPUC measure, the MBA collects in its Quarterly Mortgage Bankers Performance Report survey information on total loan production expenses that include both fixed and variable costs, such as commissions, Capacity Constraints compensation, occupancy and equipment, and other An often-made argument is that capacity constraints in the production expenses and corporate allocations. With the mortgage origination business have been particularly tight in caveat that the sample of respondents is composed of small- recent years, and that these constraints become binding when and medium-sized independent mortgage companies, the the application volume increases significantly, usually due to data indicate a modest increase in loan production expenses a refinancing wave. As a result, originators do not lower rates over the past few years and a fairly stable pattern of these as much as they would without these constraints, in order to expenses. For example, total loan production expenses curb the excess flow of applications. averaged $4,717 per loan in 2008, and $5,163 per loan in Chart 9 provides some long-horizon evidence on the 2012:Q3.43 This modest increase appears unlikely to explain potential importance of capacity constraints for profits, by the more than doubling in OPUCs over the period 2008-12. plotting our OPUC measure against the MBA application index (including both purchase and refinancing applications). The chart shows that the two series correlate quite strongly: Whenever the MBA application index increases, OPUCs tend 4.2 Industry Dynamics and Originators’ to increase, and vice-versa.44 Profits This correlation suggests that capacity constraints play an important role in generating the higher OPUCs. That said, The discussion in the previous subsection appears to indicate mortgage applications (and other measures of demand and that the higher OPUCs on regular agency-securitized loans origination activity, such as MBS issuance) were at higher levels over the period 2008-12 were not likely driven exclusively, or in the past, without OPUCs being as high as they were in 2012. even mostly, by increases in costs. As a result, the rise in OPUCs Chart 10 shows some more direct evidence on the potential during this time could reflect an increase in profits. If so, what importance of capacity constraints, by depicting the number are the potential driving forces behind such an increase? of days it takes from the initiation of a refinancing application to the funding of the loan. The chart is based on data from the
43 Source: Mortgage Bankers Association, Press Release Performance Report, various issues. The numbers cited are gross expenses, not including any 44 Over the period 2004-08, the relationship between the two series appears revenue such as loan origination fees or other underwriting, processing, or weaker than elsewhere—OPUCs appear to be on a downward trend over administrative fees. much of that time, even when applications increase.
FRBNY Economic Policy Review / December 2013 33 Chart 10 takes time to train new hires. New originators can enter the Time from Refinancing Application to Funding market, but entry requires federal and/or state licensing and (by Month in Which a Loan Is Funded) approval from Fannie Mae, Freddie Mac, and Ginnie Mae to Number of days fully participate in the origination process. To the extent that 70 HMDA training may take longer than in the past, or that approval (median) Ellie Mae delays for new entrants are longer (as anecdotally reported), 60 (average) the speed of capacity expansion may have declined compared with earlier episodes.48 Another potentially important factor 50 is that the share of third-party originations (by brokers or correspondent lenders) has decreased significantly in recent 40 years (as discussed in footnote 5). Third-party originators may, in the past, have acted as a rapid way to adjust capacity, 30 especially during refinancing waves. In sum, while capacity constraints likely contributed to the rise in OPUCs in recent 20 2003 04 05 06 07 08 09 10 11 12 years, it is unlikely that they were the only source of this rise.
Sources: HMDA (January 2003 to December 2011); Ellie Mae (August 2011 to December 2012). Notes: HMDA is the Home Mortgage Disclosure Act. HMDA data Market Concentration are restricted to first-lien mortgages for owner-occupants of one-to-four-unit houses or condos. A second popular explanation for the higher profits in the mortgage origination business is that the market is highly concentrated. It is well known that the mortgage market in
Home Mortgage Disclosure Act (HMDA), which was available only through 2011 at the time of this writing, and from the Overall market concentration alone Ellie Mae Origination Insight Report, which is only available seems unlikely to explain high profits since August 2011.45 It shows that the median (HMDA) or average (Ellie Mae) number of days it takes for an application in the mortgage business. to be processed and funded has been substantially higher since 2009 than it was in prior years.46 The processing time moves in response to the MBA application volume shown earlier; for the United States is dominated by a relatively small number instance, it reached its maximum after the refinancing wave of of large banks that originate the majority of loans. However, early 2009 and increased from less than forty days in mid-2011 as shown in Chart 11, a simple measure of market concentra- to more than fifty-five days by October 2012, as refinancing tion given by the share of loans made by the largest five or ten accelerated over this period. However, to the extent that the originators actually decreased over the period 2011-12, as a HMDA and Ellie Mae data are comparable, it does not appear number of the large players reduced their market share. Thus, that it took substantially longer to originate a refinancing loan overall market concentration alone seems unlikely to explain in 2012 than it did in early 2009, making it difficult to explain high profits in the mortgage business. This would make sense the full rise in OPUCs through capacity constraints.47 from a theoretical point of view: There is no particular reason A final interesting question is how rigid capacity why a concentrated market (but with a large number of fringe constraints may be. Current originators can add staff, but it players, and price competition) should incur large profits. Recent work by Scharfstein and Sunderam (2013) comes to a different conclusion. The authors argue that looking at 45 See www.elliemae.com/origination-insight-reports/ national market concentration may mask differential trends in EMOriginationInsightReportDecember2012.pdf. local market concentration, which matters if borrowers shop 46 The average for HMDA would be higher than the median, but would show locally for their mortgages. Using data from 1994 to 2011, the similar patterns. authors find that higher concentration at the county level is 47 It is interesting to note that the time from refinancing application to funding was significantly lower in 2003, even though application volume was much higher than it was over 2008-12. This is likely driven by tighter underwriting in 48 Additionally, existing capacity may have been diverted to defending against the recent period compared with during the 2003 refinancing boom. putbacks instead of new loan origination.
34 The Rising Gap Chart 11 Chart 12 Origination Market Concentration Weighted Average Coupons of Different Loan Types Market share 100 Percent 5.0 80 Top ten lenders Refi 105 < LTV ≤ 125 (HARP) Refi LTV > 125 (HARP) 4.5 60 Top five
40 4.0 Purchase LTV ≤ 80 Refi LTV ≤ 80 20 Top one 3.5
0 2004 05 06 07 08 09 10 11 12 3.0 Jan. Mar. May Jul. Sept. Nov. Source: Inside Mortgage Finance. 2012
Sources: Fannie Mae; Freddie Mac; eMBS. Note: The data include thirty-year fixed-rate mortgages with loan amounts less than or equal to $417,000, made to borrowers with a correlated with a lower sensitivity of refinancing and mortgage FICO score of at least 720, on owner-occupied one-unit properites. rates to MBS yields. It would be interesting to extend their analysis to 2012 to see whether their findings can help explain the increase in OPUCs in that year. We next turn to an alternative explanation for why origi- which makes servicing high-LTV loans more expensive. nators could make larger profits than in the past, namely that For these two reasons, many servicers do not offer HARP they may enjoy more pricing power on some of their borrow- refinancing for loans that they are not currently servicing, ers for reasons unrelated to concentration. or only at much worse terms. The result is that the current servicer has significant pricing power over its own high- LTV borrowers looking to refinance. HARP Refinance Loans Is there evidence that lenders can exploit this higher pricing power? The observed note rates for HARP-refinanced loans are A market segment where such pricing power may have at least consistent with this idea. As shown in Chart 12, during been particularly important is the high-LTV segment, 2012 the weighted average coupons (WACs; that is, the loan which over the past years has been dominated by note rates) on HARP loans with LTVs above 105 tended to be refinancings through HARP, originally introduced in 40-50 basis points higher than those of regular refinancing or March 2009. The introduction of revised HARP rules purchase loans.50 Banks earn higher revenues on these HARP in late 2011, often referred to as “HARP 2.0,” led to a loans than on regular loans for two reasons: given the higher significant increase in HARP activity during 2012; the note rate, they will typically sell these loans into a pool with FHFA estimates that in the second and third quarters of a 50-basis-point higher coupon, which usually commands a 2012, HARP refinancings accounted for about 26 percent price premium of around 1.5-2.0 points. Furthermore, thanks 49 of total refinance volume. HARP 2.0 provides significant to the prepayment protection offered by these pools (as a incentives for same-servicer refinancing (namely, relief borrower can only refinance through HARP once), investors from representations and warranties) that are not present are willing to pay a higher price (in the spec-pool market) than to the same extent for different-servicer refinancings. for TBA pools; this can add another 1-3 points (depending on Furthermore, even under identical representation and the coupon) to the originator’s revenue. warranty conditions, a new servicer may be less willing to add high-LTV borrowers to its servicing book, because such borrowers have a higher likelihood of delinquency, 50 We can also compare WACs on refinancings with LTV between 80 and 95 that are likely to be HARP loans (based on mortgage insurance information) with other loans in the same LTV range that are likely non-HARP loans. On average, 49 See http://www.fhfa.gov/webfiles/24967/Nov2012RefiReport.pdf. the WAC on HARP loans was about 15-20 basis points higher in that range.
FRBNY Economic Policy Review / December 2013 35 Are these higher revenues compensation for higher pricing power is not a trivial task, and we do not attempt a origination costs for HARP loans? This seems unlikely, as full analysis here. However, looking at some cross-sectional the documentation requirements for HARP loans are in fact patterns may suggest some insights. significantly lighter than for regular loans. Thus, it is likely Chart 12 shows that over 2012, the WAC on non-HARP that origination costs are lower, not higher, for HARP loans refinancing loans tended to be slightly larger than it was on relative to regular refinancings.51 purchase loans. This is somewhat surprising if one thinks that Another possibility is that high-LTV borrowers are more cash the costs of originating a refinance loan are likely lower than constrained than regular refinancers and thus require higher rebates (negative points) at origination to help cover their closing costs. While this is a possibility, it is unlikely that the difference The evidence strongly suggests that can offset a significant portion of the additional revenues, originators have been making larger especially since closing costs are likely lower than they are for regular loans (thanks, for example, to appraisal waivers).52 profits on HARP loans than on regular Finally, for reasons discussed above, the value of base loans, by being able to exploit their servicing on HARP loans may be significantly lower than that pricing power. for non-HARP loans with lower LTVs. Even if we assume that the multiple on base servicing drops from 5x to 0x, however, this would only account for 1.25 points, while, as noted above, revenues are 2.5-5.0 points higher. Furthermore, because those of a purchase loan. In addition, comparing WACs over a HARP borrowers are expected to prepay slowly, the cash flow longer time period (not shown), it is the case that the positive stream from servicing is in fact more valuable than for regular gap in WACs between purchase and refinancing loans only loans, offsetting part of the higher servicing cost. Also, the started emerging in 2010 (and has remained there since); over expected servicing cost for current servicers declines when the period 2005-09, average monthly WACs on refinancing loans are refinanced under HARP, as borrowers are less likely loans were mostly either equal to or below those on purchase to default after the note rate declines (see Tracy and Wright loans.53 However, the WAC divergence could potentially be [2012] and Zhu [2012]). explained by purchase borrowers paying more points than Thus, the evidence strongly suggests that originators have refinancers; this could be, for instance, because they expect to been making larger profits on HARP loans than on regular stay in the mortgage longer or because of tax incentives.54 loans, by being able to exploit their pricing power. One would expect this explanation, if true, to hold across all lenders. However, looking at lender-specific differences in WACs reveals a large variation across lenders. The two panels of Chart 13 show the monthly average WAC for the sixteen Non-HARP Mortgages largest lenders over 2012 (in terms of number of loans sold to the GSEs), for purchase and refinancing loans separately. The next question is whether similar pricing power could have We also plot separately the average for all other (smaller) contributed to the rise in our OPUCs on regular (non-HARP) sellers (the thicker lines). We include only thirty-year fixed- loans that seems not fully explained by capacity constraints, rate loans with FICO scores of 720 and higher, and LTVs of as discussed above. While lenders may have pricing power 80 or lower, made to single-unit owner-occupiers in order to over their HARP borrowers, it is much less clear whether reduce potential disparities due to differential LLPAs.55 such pricing power may also exist for “regular” loans. Pricing power could arise, for instance, from customers’ impediments (actual or perceived) to shop around, an unwillingness of other firms to compete, barriers to entry for new competitors, 53 or a combination of these. Directly measuring originators’ This statement is based on loan-level data from Freddie Mac only, as the Fannie Mae data only became available in 2012. 54 Points paid in cash are fully tax deductible for purchase mortgages in the 51 Also, the loans with FICO scores of 720 or above that we include in the year the loan is closed. For refinancing mortgages, the deduction is instead chart are not subject to loan-level price adjustments under HARP. spread evenly over the term of the mortgage (for example, thirty years),
52 except if the loan is paid off early, in which case all unused deductions can Related to this point, it is not the case that HARP note rates are higher be taken in the year the loan is paid off. See, for example, www.irs.gov/ because principal amounts are lower than for regular refinancings (as the same publications/p936/ar02.html#en_US_2011_publink1000229936. fixed closing cost being rolled into the rate will require a larger rate increase for lower principal amounts); controlling for loan amount in a regression basically 55 These calculations are based on the complete set of loan-level disclosures leaves the estimated differences across loan categories unchanged. for pools issued in 2012 by Fannie Mae and Freddie Mac.
36 The Rising Gap Chart 13 in at least one month, and, for six of them, that is the case for Dispersion in Weighted Average Coupons across at least six out of twelve months.56 In principle, this observed Sellers to Fannie Mae and Freddie Mac, 2012 Percent price dispersion is certainly not inconsistent with the market being competitive; however, under this null hypothesis, it is surprising that the dispersion is so much larger for refinancing Panel A: Purchase loans loans than for purchases. 4.75 As discussed above, during 2012 the HARP program 4.50 gained significant momentum for high-LTV refinances. A perhaps lesser-known fact is that there exist GSE streamline 4.25 Average of smaller sellers refinancing programs also for non-HARP loans (with LTV less than 80), with the same cutoff date for eligible mortgages 4.00 (which must have been delivered to one of the GSEs prior to
3.75 May 31, 2009). Streamlined refinancing, when done through the institution that currently services the loan, relieves the 3.50 lender from representation and warranties relating to the borrower’s creditworthiness and home value, while a different- 3.25 servicer refinancing requires more extensive underwriting Panel B: Refinancing loans of the new loan. As a consequence, for borrowers eligible for 4.75 a streamlined refinancing, there is an advantage to staying Average of with the same servicer/lender, as doing so will reduce the 4.50 smaller sellers documentation the borrower is required to submit. This, 4.25 in turn, again creates some pricing power for the current servicer (although likely less so than for high-LTV loans). 4.00 The population of loans in fixed-rate GSE pools originated prior to June 2009 is large: As of December 2012, about 3.75 $1.1 trillion of loans were in such pools, relative to an
3.50 overall Fannie Mae/Freddie Mac fixed-rate universe of about $3.8 trillion. During 2012, about 52 percent of all prepayments 3.25 came from pools issued prior to June 2009.57 Therefore, if Jan. Mar. May July Aug. Nov. lenders have pricing power over the refinancings of these loans, this could be a nontrivial contributor to OPUCs. Sources: Fannie Mae; Freddie Mac; eMBS. Is there evidence that such pricing power could explain Note: The data include loans with a FICO score of 720 or higher, the dispersion in refinancing WACs? Unfortunately, unlike an LTV of 80 or lower, an amount less than or equal to $417,000, on owner-occupied single-unit properties, and only for months in for HARP loans, there is no way for us to observe in the data which a seller made at least 100 sales. whether a refinancing was streamlined or not. However, we can look at variation across lenders in the fraction of their servicing portfolio that could potentially be refinanced in a streamlined manner (that is, loans in pools issued prior to June 2009) and correlate this figure with the average Panel A of the chart shows that purchase WACs across WAC of the lenders’ non-HARP refinance loans over 2012. sellers were quite homogeneous—with the exception of a Chart 14 shows that there is indeed a positive correlation couple of outliers, most lender WACs lie within a range of between the two: The lenders that had a large fraction approximately 10 basis points. This is consistent with the idea of potentially streamline-eligible loans in their servicing that the purchase mortgage market is quite competitive, as presumably many borrowers shop around (perhaps with the 56 With the exception of one of these six lenders, the monthly number of sales help of their realtor). of refinancing loans always exceeds 500 loans, meaning that these averages Panel B reveals a much larger dispersion for refinancing are unlikely to be driven by small-sample noise. Additionally, as above, the result of large WAC dispersion across lenders for refinance loans remains loans. In particular, while a number of sellers remain con- basically unchanged if loan characteristics such as loan amount are added as centrated around the thicker line representing the average of explanatory variables in a regression framework. smaller players, eight of these large lenders sold loans with 57 These prepayments include refinancings as well as the loan simply getting WACs that are 15 basis points or more above the thick line paid off (for instance, due to the borrower moving).
FRBNY Economic Policy Review / December 2013 37 Chart 14 5. Conclusions Weighted Average Coupons on Regular (Low LTV) Refinance Loans Against Fraction of Servicers’ Portfolio Eligible for Streamline Refinancing The widening gap between primary and secondary mortgage rates over the period 2008 to 2012 was due to a rise in orig- inators’ profits and unmeasured costs, or OPUCs, as well as Weighted average coupons of non-HARP refis relative to smaller increases in g-fees. The magnitude of the OPUCs is influenced sellers over 2012 (averaged across months, in percent) 0.5 by MBS prices, the valuation of servicing rights, points paid by borrowers, and costs such as those from loan putbacks and 0.4 pipeline hedging. 0.3 The rise in OPUCs was mainly driven by higher MBS prices, which were not offset by corresponding increases 0.2 in measurable costs. Conversely, a decline in the value of 0.1 mortgage servicing rights may have reduced OPUCs to some extent, and thus contributed to the widening primary- 0 secondary spread. Among harder-to-measure costs, we find -0.1 that expected putback costs and pipeline hedging likely did not cause a significant portion of the rise in OPUCs. Absent -0.2 0.30.4 0.50.6 0.7 increases in other costs that we cannot measure well, such as operating costs, the rise in OPUCs reflected an increase in Share of November 2011 servicing portfolio in HARP- or streamline-eligible pools originator profits. While market concentration alone does not seem to explain the rise in these profits, capacity constraints Sources: Fannie Mae; Freddie Mac; eMBS. do appear to have played a significant role. We also provide Notes: HARP- or streamline-eligible pools are pools issued prior to June 2009. The data include only sellers/services with servicing evidence suggesting that originators have enjoyed pricing portfolios with more than $1 billion of HARP- or streamline-eligible power on some of their borrowers looking to refinance, due to pools in November 2011. Non-HARP weighted-average coupons are borrowers’ switching costs. calculated on loans with a FICO score of 720 or higher, an LTV of 80 or lower, an amount less than or equal to $417,000, on owner- Going forward, it will be interesting to study the extent to occupied single-unit properties. which interest rate dynamics, capacity expansions, new entry, changes in regulations, and (in the longer term) housing finance reform will affect the pass-through from secondary to primary markets. As illustrated in this article, a number of factors deter- portfolio at the end of 2011 tend to be those that originated mine this pass-through, and it will therefore be important for refinance loans with the highest WACs on average over policymakers and market participants alike to further improve 2012 (that is, those that are above the thick line in panel B the measurement and understanding of these factors. of Chart 13). This result is consistent with (though certainly not proof of) originators taking advantage of their pricing power over streamline-eligible borrowers.
38 The Rising Gap References
Bhattacharya, A. K., W. S. Berliner, and F. J. Fabozzi. 2008. “The Vickery, J., and J. Wright. 2013. “TBA Trading and Liquidity in Interaction of MBS Markets and Primary Mortgage Rates.” the Agency MBS Market.” Federal Reserve Bank of New York Journal of Structured Finance 14, no. 3 (fall): 16-36. Economic Policy Review 19, no. 1 (May): 1-19.
Scharfstein, D. S., and A. Sunderam. 2013. “Concentration in Zhu, J. 2012. “Refinance and Mortgage Default: An Empirical Mortgage Lending, Refinancing Activity, and Mortgage Rates.” Analysis of the HARP’s Impact on Default Rates.” Unpublished NBER Working Paper no. 19156, June. paper, Federal Home Loan Mortgage Corporation. Available at www.ssrn.com/abstract=2184514. Tracy, J., and J. Wright. 2012. “Payment Changes and Default Risk: The Impact of Refinancing on Expected Credit Losses.” Federal Reserve Bank of New York Staff Reports, no. 562, June.
The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York, the Federal Reserve Bank of Boston, or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.
FRBNY Economic Policy Review / December 2013 39
Rajashri Chakrabarti and Sarah Sutherland
Precarious Slopes? The Great Recession, Federal Stimulus, and New Jersey Schools
• Despite the significance of the Great “The Legislature shall provide for the maintenance and Recession’s impact on the economy, virtually support of a thorough and efficient system of free public no research exists on how schools were schools for the instruction of all the children in the State affected. between the ages of five and eighteen years.” New Jersey State Constitution, Article V, Section 4 • This article fills the gap by investigating the effect of the downturn and the federal “[T]he New Jersey State Constitution requires the Gov- stimulus on New Jersey school finance ernor to take care that the laws of this State be faithfully patterns. executed…including ensuring compliance with the constitutional mandate that a balanced State budget be • The authors provide strong evidence of maintained.” downward shifts in total school funding and State of New Jersey Executive Order No. 14, February 11, 2010 expenditures, relative to trend, following the recession.
• The $2 billion-plus in stimulus earmarked 1. Introduction for New Jersey limited damage to budgets; it also helped preserve funding levels in The relevance of the investment in the education of children instructional expenditure, the category most to human capital formation and economic growth is well closely related to student learning. established in economic research.1 Surprisingly, then, one important component of this topic has been overlooked in the • The study offers insight into how school literature: the impact of recessions on education. districts fare during downturns and serves as The Great Recession was marked by a downturn in housing a useful guide for future policy decisions. prices, employment, and business activity, each of which
1 See Barro (1991), Benhabib and Spiegel (1994), Becker (1994), and Hanushek and Woessmann (2007).
Rajashri Chakrabarti is an economist at the Federal Reserve Bank of The views expressed are those of the authors and do not necessarily reflect New York; Sarah Sutherland is a research associate at the Bank. the position of the Federal Reserve Bank of New York or the Federal Re- Correspondence: [email protected] serve System.
FRBNY Economic Policy Review / December 2013 41 contributed to reduced tax revenues and larger budget gaps.2 We also delve deeper into how the different components of These shortfalls had a deleterious effect on state and local gov- expenditures are affected. While instructional expenditure per ernments’ ability to fully fund schools. While a sparse literature pupil declined in 2009 relative to trend, there is no evidence of investigates the impact of the severe downturn on other sectors any decrease in 2010. Thus, the federal stimulus funding may of the economy, there is virtually no research on the effects of have been successful in preventing declines in instructional the Great Recession, or past recessions, on schools. Our paper expenditure, a category including teacher salaries and class- starts to fill that void by examining how school finances in New room expenses that most directly affect student learning. Jersey were affected by the onset of the recession and the federal The patterns for instructional support per pupil and stu- stimulus funding that followed. Using rich panel data capturing dent services per pupil are similar to those for instructional a multitude of school finance variables, we apply a trend-shift expenditure. But other noninstructional categories such as analysis to study how the Great Recession and federal stimulus transportation and utilities and maintenance (“utilities”) show affected the level and composition of funding and expenditures declines relative to trend in 2009 and 2010. Median teacher sal- in New Jersey school districts. Our findings offer insight into aries show a positive shift in both years, while median teacher schools’ financial situations during recessions and can assist in experience also increased. These patterns are consistent with future policy decisions. an increase in layoffs of less experienced teachers, which would Demonstrating concern for safeguarding schools during shift the teacher salary distribution to a higher range. a recession marked by pervasive budget cuts, the federal Despite these statewide patterns, there is considerable het- government designated the largest portion of its planned erogeneity by poverty level, metropolitan area, and urban sta- economic stimulus package for public education. In February tus. Specifically, “high-poverty” districts sustained larger falls 2009, Congress passed the American Recovery and Rein- relative to trend in nearly all expenditure categories compared vestment Act (ARRA), an economic stimulus package which with their “medium-poverty” and “affluent” counterparts. The provided $840 billion in new spending, with $100 billion des- metropolitan area of Edison fared best in terms of preserving ignated for public education. Of this $100 billion, New Jersey instructional expenditure and most noninstructional expen- was allocated $2.23 billion.3 ditures. Finally, rural districts fare best across most categories, In addition to studying the overall impact of the reces- while urban districts experience the largest resource declines. sion on schools, we examine whether the effects varied by a This paper builds on the existing literature relating to school district’s poverty level, metropolitan area, and urban status. funding in general (Baker 2009, Bedard and Brown 2000, Betts The analysis yields some interesting results. There is strong 1995, Feldstein 1978, Gordon 2004, Rubenstein et al. 2007, and evidence of a downward shift—relative to trend—in both rev- Stiefel and Schwartz 2011), and the literature on New Jersey enue and expenditure following the recession in New Jersey. school finance (Firestone et al. 1994 and Firestone, Goertz, and Federal stimulus seemed to have helped in 2010; while reve- Natriello 1997). While these authors study school funding pat- nue and expenditure still fell (relative to trend), the declines terns broadly, our paper is one of the first to examine whether were somewhat smaller than in 2009. (We refer to school years a recession affects school finance patterns, and what difference by the year corresponding to the spring semester.) federal stimulus funding can have on the trends. While total funding to schools declined, the various com- It is worth noting that we view our findings as strongly ponents of aid did not experience symmetric changes. State aid suggestive, but not necessarily causal. We employ a trend- per pupil fell in both years after the recession (relative to trend), shift analysis, so theoretically if there were common shocks as did local funding per pupil. However, the percentage decline in the two years following the recession that could affect our in state aid per pupil far exceeded the corresponding decline financial variables, they would bias our estimates. We conduct in local funding per pupil, especially in the second year after a comprehensive analysis of potential confounding factors the recession. In contrast, there was an upward shift in federal during this period that might bias our results (see section 4). aid per pupil in 2010 following the introduction of the ARRA This analysis helps us interpret the results, frame our perspec- funds. These changes marked an important shift in the relative tive, and put bounds on the recession-impact estimates. reliance of schools on federal, state, and local funding. Finally, the Great Recession was not a marginal shock; rather, it was a highly discontinuous shock. Therefore, even if there were other small shocks during these two years, they 2 See Gerst and Wilson (2010) and Deitz, Haughwout, and Steindel (2010) for would be dwarfed by a downturn as large as the Great Reces- an analysis of budget gaps nationally and in New Jersey. sion, adding further confidence to our results. 3 The $2.23 billion figure comes from information provided by the New Jersey Studying school funding during this period is of para- Department of Treasury’s Office of Management and Budget and represents the total ARRA appropriation for the state. mount importance because schools have a fundamental role
42 Precarious Slopes in educating children and fostering human capital forma- To remedy these depletions, Congress passed ARRA in tion and growth. Any adverse effect on schools and student February 2009, an economic stimulus package that provided learning can have potentially deleterious effects on human $840 billion in new spending, with $100 billion designated capital formation and by extension the nation’s future. Our for public education. Districts were directed to use the ARRA paper provides insight into how school districts fared during funds to save and create jobs and to boost student achieve- the financial downturn and promises to both improve our ment. The requirements specified that 81.8 percent of the understanding of schools’ financial situations under duress as stabilization funds go toward the support of public education, well as aid future policy decisions. and that states must restore public education funding in fiscal years 2009, 2010, and 2011 to the greater of the fiscal year 2008 or fiscal year 2009 level. Of the $100 billion earmarked for public education nation- 2. Background ally, New Jersey received $2.23 billion. The largest portion of New Jersey’s appropriation was used to implement the state’s school funding formula, and these funds were spent by the end of the 2010 school year. 2.1 Overview of the Period’s Economic Climate and Education Policies 2.2 Overview of New Jersey’s Education State and local governments in the United States experienced History and Programs significant fiscal stress as a result of the Great Recession. The downturns in housing prices, employment, income, and busi- In January 2008, the School Funding Reform Act (SFRA) was ness activity each contributed to lower tax revenues and larger approved by the New Jersey legislature. This Act was the state’s budget gaps. first official change to its school funding formula since 1996 Local governments have, in the past, relied heavily on prop- and was the product of five years of development by the state’s erty taxes, which in the first half of the decade were supported Department of Education. The formula called for a 7 percent by a booming housing market. Housing prices in the United increase in state funding for K-12 education in the 2009 States had been increasing at an annual average rate of 7.8 per- school year. The recession officially began in December 2007, cent between 2000 and 2006, but as delinquencies and foreclo- and since governments finalize their budgets in the spring sures began to rise, home prices declined at an annual average prior to the budgeted year, the education budgets for the 2009 rate of 3.9 percent during the recession quarters.4 Demonstrat- school year were the first to be affected. Despite the start of ing an even greater swing than the rest of the country, housing the recession, the amount required by the new SFRA formula prices in New Jersey were increasing at a brisk average of 11.6 was fully met in the 2009 school year, and 2010 budgets were percent per year between 2000 and 2006, and fell by an average also prepared using the formula. 4.9 percent per year in the recession quarters. Housing price Midway through 2010, however, the toll of the recession declines are one of the many contributors to the decline in state forced some changes. Revenue streams were projected to be and local revenues during the Great Recession. $2.2 billion lower than what was necessary to cover the state’s State governments also took in less revenue as unem- budget deficit. Given New Jersey’s constitutional mandate to ployment spikes reduced income taxes collected and lower maintain a balanced budget, education funding was reduced consumption generated fewer sales taxes. New Jersey also midyear.6 The funding caps for district aid were lowered, and relied heavily on the financial industry to provide an increas- many districts received less state aid than budgeted and less ing portion of its revenues, but with the recession hitting the aid than required under the SFRA formula. finance sector hard, the state’s budget gap grew.5
4 For all figures related to home prices, we use the annualized four-quarter price change in the Federal Housing Finance Agency (FHFA) House Price
Index averaged over the specified time period. Recession quarters are based 6 on the National Bureau of Economic Research’s definition. In February 2010, a fiscal emergency was declared in New Jersey due to the projected $2.2 billion budget deficit for fiscal year 2010 and a range of cuts 5 Deitz, Haughwout, and Steindel (2010). were made to ensure compliance with the state’s balanced-budget mandate.
FRBNY Economic Policy Review / December 2013 43 Box 1 3. Data Definitions of Expenditure Components
We developed a rich set of panel data combining annual data Instructional at the school district level from multiple sources. The data Instructional Expenditures set covers 572 New Jersey districts for the school years 1999 All expenditures associated with direct classroom instruction: through 2010.7 Most of the data were obtained from the New teacher salaries and benefits; classroom supplies. Jersey Department of Education’s Office of School Finance. Noninstructional We supplemented this data set with school finance data from the National Center for Education Statistics (NCES) School Instructional Support Finance Survey (F-33) as well as data from the U.S. Census All support service expenditures designed to assess and improve Bureau. Nonfinance data were obtained from the New Jersey students’ well-being: food services, educational television, Department of Education’s Office of Data, Research, Evalua- library, and computer costs. tion, and Reporting, NCES’s Common Core of Data (CCD), Student Services and the U.S. Bureau of Labor Statistics. Psychological and health services; school store. The information includes data on total funding, total Utilities and Maintenance expenditure, and debt outstanding, as well as data on indi- Heating, lighting, water, and sewage; operation and maintenance. vidual components of total funding and expenditure. On the Transportation expenditure side, for example, the figures include spend- Total expenditure on student transportation services. ing on instruction, instructional support, student services, Student Activities transportation, student activities, and utilities. (See Box 1 Co-curricular activities: physical education, publications, clubs, for definitions of these variables.) We also obtained data on and band. federal aid, state aid, local funding, property tax revenue, and data on median salaries and median years of experience for teachers and administrators.8 Nonfinance data include district-level data on various socioeconomic and demographic variables, including enroll- ment, racial composition, and percentage of students eligible We categorize by poverty level and urbanicity based on for free or reduced-price lunches. All funding and expenditure the 2008 levels in an effort to capture pre-recession mea- variables are analyzed on a per-pupil basis using each school sures. Poverty level is defined by the percentage of students year’s average daily enrollment. eligible for free or reduced-price lunches. For example, the Heterogeneity breakdowns are performed by metropoli- districts falling above the seventy-fifth percentile in terms of tan division (MD), poverty level, and “urbanicity.” MDs are the percentage of students eligible for free or reduced-price groupings of counties or equivalent entities defined by the lunches are identified as high poverty, those in the bottom U.S. Census Bureau. They are smaller than metropolitan twenty-fifth percentile as affluent, and districts falling between statistical areas but contain a population of at least 2.5 mil- these percentiles as medium poverty. Urbanicity designations lion.9 Heterogeneity breakdowns for metropolitan areas of rural, urban, or suburban are defined using the NCES CCD 11 include the four largest New Jersey MDs: New York-White classifications. Plains-Wayne, Edison-New Brunswick, Newark-Union, and To account for inflation, all expenditure and aid data were Camden (see Map 1).10 adjusted to 2010 dollars using annual values of the consumer price index for all urban consumers.12
7 See Gerst and Wilson (2010) and Deitz, Haughwout, and Steindel (2010) for an analysis of budget gaps nationally and in New Jersey. 8 The $2.23 billion figure comes from information provided by the New Jersey
Department of Treasury’s Office of Management and Budget and represents 11 the total ARRA appropriation for the state. As a point of reference, in the 2007-08 school year, districts at the high poverty level had 27.7 percent of their students receiving free or reduced- 9 We use ArcGIS mapping technology and U.S. Census Bureau data to define a price lunches, while the affluent districts had 5.1 percent receiving the same. district’s metropolitan division. 12 Districts in urbanized areas or urban clusters less than thirty-five miles from 10 The New York-White Plains-Wayne metropolitan statistical area includes urbanized areas are categorized as urban. Territories outside principal cities and counties in both New York and New Jersey districts. Since it comprises a very in urbanized areas represent the suburban districts. NCES uses the U.S. Census populated part of the state of New Jersey, we include it here. Bureau definition of rural territory based on the level of land developed.
44 Precarious Slopes M To quantify the change in each finance variable, we also Four Largest New Jersey Metropolitan Divisions compute percentage shifts that are obtained by expressing the
shifts α( 2 and [α 2 + α 3]) from our specification as percentages of the pre-recession (2008) base of the corresponding financial
New York- variable (Y it). This pre-recession base is simply the average of White Plains- each in the 2008 school year. Recall that local, state, and fed- Wayne, NY-NJ eral governments finalize their budgets in the spring prior to the budgeted year. More specifically, the budgets for the 2008 Newark-Union, NJ-PA school year were finalized in the spring of 2007, before the re- cession officially began in December 2007, and before decision makers were aware of the impending downturn. Therefore, 2008 is taken as the last pre-recession year in this paper. These percentage effects allow for an easier interpretation
and are more informative than the coefficientsα ( 2 and α 3 ) alone, since they suggest the size of the effects and facilitate Edison- New Brunswick, NJ comparison between the shifts in the various financial vari- ables. In our discussion, we focus on two percentage shifts: the 2009 percentage shift immediately following the recession α 2 Camden, NJ (calculated as ______ ) for each finance variable (Y ) and pre-recession base it α 2 the percentage shift in 2010 (calculated as ______ ) for pre-recession base each finance variable Y( it). The first percentage shift captures the effect of the recession in 2009 and the second captures the combined effect of the recession and the federal stimulus in 2010. Note that if there were common shocks in the two years following the recession affecting our financial variables, our estimates of the recession and stimulus effects outlined above would be biased. Understanding these potentially confound- ing factors is essential for interpreting the results. Therefore, we conduct a thorough analysis of them during the period. First, while interpreting the shift at the onset of the reces- Source: Authors’ representations using U.S. Census Bureau shape les. sion, we consider the implementation of the new school fund- Note: Our analysis focuses on the four largest New Jersey Metropolitan ing formula under the SFRA. The formula called for a 7 percent Divisions identi ed on this map. increase in total state funding for K-12 education in the 2009 school year. Since state aid constitutes nearly half of the gen- eral-formula aid to districts, the new funding formula should be considered a positive shock not only to state aid, but also to 4. Interpretation of Post-Recession total funding, total expenditure, and the components of total Effects expenditure. Since the shock is in the opposite direction of the recession shock, any negative shift in the school-finance The goal of this paper is to investigate whether the Great Re- variables in 2009 would be above and beyond the expected cession and the federal stimulus funding period that followed positive effect of the new funding formula. Therefore, it is safe are associated with shifts in New Jersey education financing. to say that negative shifts (if any) in the variables in 2009 are We conduct a trend-shift analysis and use the specification underestimates of the recession effects, though in the correct described in Box 2 to analyze these effects. The intuition direction. In contrast, positive shifts could mean that the effect behind using this methodology is that school finances would of the SFRA increase surpassed the effect of the recession or be expected to continue to grow at the pre-recession trend had that the recession did not have much of a negative effect. Second, while interpreting the 2010 stimulus shift, we there been no recession. Thus, post-recession effectsα ( 2 and consider the impact of the midyear cuts to the SFRA formu- [α 2 + α 3] in Box 2) are captured by shifts from this trend both in 2009 and 2010.
FRBNY Economic Policy Review / December 2013 45 la.13 With the negative effect on state aid funding, we would structional categories such as transportation and utilities show expect a dampening effect on the positive shock from the either a flattening or a decline in the years of interest. Spend- ARRA federal stimulus. Note, however, that these cuts only ing on student activities seems to have remained on trend. came at midyear and did not affect schools’ planned budgets Funding for instructional support and student services shows or their expenditures in the first half of the school year. Any signs of flattening in the first year, then a reversal in 2010. positive additional effects in 2010 (over 2009 effects) could Teacher and administrator salaries (Chart 3) show an upward therefore be regarded as underestimates of the stimulus effect. shift in the post-recession period, at least in 2010. In the next Additional negative effects in 2010, however, could mean that section, we investigate whether these patterns continue to the recession effects (including the midyear cuts) dominated. hold in a more formal trend-shift analysis. As noted in our overview of New Jersey’s education programs Tables 1-6 present results from the estimation of our specifi- in section 2.2, these midyear cuts in 2010 were driven by cation described in Box 2. Each of these tables is structured the budget shortfalls brought about by the recession. In that sense, same way. The top section of each panel presents the percent- the 2010 shift would still capture the combined effects of the age shifts, with the first row capturing the percentage shift in α 2 recession and stimulus funding. 2009 (calculated as ______ ), the second row capturing pre-recession base α 2 + α 3 the percentage shift in 2010 (calculated as (______ ), pre-recession base and the third row showing the pre-recession base of the
corresponding school finance variable as (Y it ). The bottom 5. Results section of each panel presents the regression estimations from which the percentage shifts were derived. Referring back to the equation in Box 2, “Trend” in this panel corresponds to
α1, “Recession” to α 2, and “Stimulus” to α 3. Our discussion of results will focus on the shifts. For easier comparability and 5.1 Overall Patterns a visual representation, the same percentage shifts are also illustrated in the corresponding histograms in Chart 4 and Using all 572 New Jersey districts in our data set, Chart 1 Charts 6-9. shows the general statewide trends in per-pupil expenditure As Table 1, panel A, and Chart 4 show, both total expendi- and funding, as well as the change over time in federal, state ture and total funding experience downward shifts relative to and local contributions to total funding. The dashed line trend in 2009, signifying the negative effect of the recession. represents the last pre-recession year of 2008 and the x-axis Again, note that these effects are likely underestimates of the represents the spring term of the school year with the last data corresponding recession effects since the change in the SFRA point showing 2010. Both total expenditure and total funding funding formula had a positive effect on overall school aid. show declines in 2009. Despite a slight increase in 2010, the Declines are evident in 2010 as well, but they are somewhat levels for total expenditure and total funding did not return to more modest, at least for total expenditure per pupil. pre-recession per-pupil levels. Federal aid increased 1 percent As we expected, federal aid per pupil shows a sharp upward between 2008 and 2009, and then jumped 35 percent from shift relative to trend in 2010, coinciding with the infusion of 2009 and 2010, the year of the federal stimulus funding. Dis- federal funds from the ARRA stimulus. In contrast, state aid trict reliance on federal aid spiked 32 percent between 2009 per pupil declines from trend in both years after the recession. and 2010, while reliance on state aid dropped 16 percent. But Recall from our earlier discussion that the SFRA led to an local funding and property tax declined and flattened out in upward shift in state aid per pupil in 2009, so the decline that the post-recession period. Relative reliance on local fund- year is likely an underestimation of the true recession effect. ing actually shifted upward from pre-recession levels, while Digging deeper, we find that although the funding targets set reliance on state aid declined. (The former is due to a sharp by the SFRA formula were achieved in 2009, state-level cuts in decrease in state aid.) Finally, Chart 1 shows a flattening of categories outside the formula, such as pension funding, led to enrollment in the post-recession period. a negative shift in overall state aid. Chart 2 analyzes compositional changes in expenditures. Historically, a significant portion of state aid has been Instructional expenditures show evidence of flattening in distributed to the New Jersey Teachers’ Pension and Annui- 2009, with the pattern reversing in 2010. In contrast, nonin- ty Fund (TPAF). Allocations are not stipulated in the SFRA formula, so in 2009, when the recession began depleting 13 Midway through the 2009-10 school year, the funding caps for district aid revenue flows, pension funding was cut dramatically. Chart 5 were lowered, and many districts received less state aid than was budgeted and less aid than was required under the SFRA formula. shows the trends in total state aid, TPAF funding, and state aid
46 Precarious Slopes Chart 1 Trends in School Revenue and Expenditures during the Great Recession
Dollars Dollars Dollars 22,000 Total Expenditures per Pupil 24,000 Total Funding per Pupil 1,000 Federal Aid per Pupil
20,000 22,000 800 20,000 18,000 600 18,000 16,000 400 16,000
14,000 14,000 200
12,000 12,000 0 1999 02 04 08 10 1999 02 04 06 08 10 1999 02 04 06 08 10 Dollars Dollars Dollars 10,000 State Aid per Pupil 12,000 Local Aid per Pupil 10,000 Property Taxes per Pupil
8,000 10,000 8,000 8,000 6,000 6,000 6,000 4,000 4,000 4,000
2,000 2,000 2,000
0 0 0 1999 02 04 06 08 10 1999 02 04 06 08 10 1999 02 04 06 08 10 Percent Percent Percent 4.0 Contribution from Federal Aid 35 Contribution from State Aid 54 Contribution from Local Aid 3.5 30 52 3.0 25 2.5 20 50 2.0 15 1.5 48 10 1.0 46 0.5 5 0 0 44 1999 02 04 06 08 10 1999 02 04 06 08 10 1999 02 04 06 08 10 Millions 14.0 Total Students 13.5
13.0
12.5
12.0
11.5 1999 02 04 06 08 10
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary and Taxpayers’ Guide to Education Spending. Note: Years denote spring terms.
FRBNY Economic Policy Review / December 2013 47 Chart 2 Trends in School Revenue and Composition of Expenditures during the Great Recession
Dollars Dollars Dollars 10,000 Instructional Expenditures per Pupil 2,500 Instructional Support per Pupil 2,000 Student Services per Pupil
8,000 2,000 1,500
6,000 1,500 1,000 4,000 1,000
500 2,000 500
0 0 0
300 Student Activities per Pupil 1,000 Transportation per Pupil 2,000 Utilities and Maintenance per Pupil
250 800 1,500 200 600 150 1,000 400 100 500 50 200 0 0 0 1999 02 04 06 08 10 1999 02 04 06 08 10 1999 02 04 06 08 10
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary and Taxpayers’ Guide to Education Spending. Note: Years denote spring terms.
C less TPAF funding. The vertical line represents the immediate Median Salary for Teachers and Administrators pre-recession school year. While total state aid declined be- tween 2008 and 2009, in large part due to the decline in TPAF Dollars Dollars funding, state aid less TPAF increased. So while the SFRA in- 65,000 150,000 sulated other parts of state aid, the recession adversely affected 62,500 140,000 TPAF funding in 2009, which, in turn, negatively affected total 60,000 Teacher salary 130,000 state aid per pupil. Scale These patterns are also reflected in a formal trend-shift 57,500 120,000 Administrative salary analysis of the state aid components. (For brevity, correspond- Scale 55,000 110,000 ing estimates are not reported, but are available on request.) 52,500 100,000 The situation in 2010, however, is different. Although the state 50,000 90,000 budgets for 2010 were established using the SFRA formu- 2001 02 03 04 05 06 07 08 09 10 la, revenue streams that year were less than expected. In an Source: Authors’ calculations, using the New Jersey Department unprecedented move, the funding formula was revamped of Education’s Report Card data. significantly at midyear. The funding caps for district aid Notes: Years denote spring terms. Y-axis gures represent were lowered, and many districts received considerably less averages of all district medians. Data are ination adjusted. state aid than was budgeted and less aid than was required under the SFRA funding formula. Indeed, Chart 5 shows that declines are evident in total state aid as well as in state aid less TPAF. Results for 2009 and 2010 suggest that the Great Reces- sion had a marked negative effect on state aid to districts.
48 Precarious Slopes? Table 1 Patterns in Revenue and Expenditures per Pupil during the Financial Crisis and Fiscal Stimulus Period
(1) (2) (3) (4) (5) (6) Total Expendi- Total Funding Federal Aid State Aid Local Aid Property Taxes Panel A tures per Pupil per Pupil per Pupil per Pupil per Pupil per Pupil
Percentage shift in 2008-09 -9.51*** -12.68*** -17.5*** -4.02*** -3.36*** -2.81*** Percentage shift in 2009-10 -8.48*** -12.58*** 13.02*** -18.46*** -2.66*** -1.74**
Pre-recession base 20,180 23,460 565 6,220 11,539 11,093
Trend 567.6*** 694.9*** 15.3*** 116.4*** 349.16*** 326.33*** (25.0) (38.9) (2.0) (6.3) (10.9) (10.2) Recession -1,919.0*** -2,974.1*** -98.8*** -250.2*** -387.61*** -311.28*** (161.0) (295.1) (16.6) (48.8) (66.3) (63.2) Stimulus 208.2 24.9 172.4*** -897.9*** 80.67 118.39 (195.2) (338.9) (18.2) (59.8) (93.6) (89.9)
Observations 6,753 6,753 6,753 6,753 6,753 6,495 R2 0.576 0.511 0.829 0.936 0.869 0.884
(7) (8) (9) (10) Contribution Contribution Contribution Total Panel B from Federal Aid from State Aid from Local Aid Students
Percentage shift in 2008-09 -10.81*** 4.01*** 7.48*** -3.56*** Perecentage shift in 2009-10 20.71*** -10.82*** 7.58*** -4.26***
Pre-recession base 2 27.8 51.3 2,384.9
Trend 0.01 -0.27*** 0.10* 14.4*** (0.0) (0.0) (0.1) (1.5) Recession -0.26*** 1.12*** 3.84*** -85.0*** (0.1) (0.2) (0.4) (14.3) Stimulus 0.76*** -4.12*** 0.05 -16.4 (0.1) (0.3) (0.5) (18.9)
Observations 6,759 6,759 6,759 6,753 R2 0.799 0.926 0.784 0.995
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary and Taxpayers’ Guide to Education Spending.
Notes: Robust standard errors are in parentheses. All regressions control for racial composition and percentage of students eligible for free or reduced-price lunch, and include school district fixed effects. Pre-recession base is expressed in 2010 constant dollars.
***Statistically significant at the 1 percent level. **Statistically significant at the 5 percent level. *Statistically significant at the 10 percent level.
As would be expected given the recession’s shock to hous- Table 1, panel B, illustrates percentage shifts in federal, ing prices and local revenue streams, local funding per pupil state, and local contributions to total school funding. The and property taxes per pupil show negative shifts relative to patterns reveal that the above changes led to districts relying trend in both years after the recession (Table1 and Chart 4). less on state aid and, instead, becoming largely funded more It follows from the above analysis that without the support of by federal aid in the 2010 school year. Thus, the Great Re- the federal stimulus in 2010, total aid to districts would have cession and the associated infusion of funds from the federal declined even further from their depleted levels. stimulus package seem to have led to a compositional shift in
FRBNY Economic Policy Review / December 2013 49 Box 2 Empirical Strategy is in spite of the 2010 cuts to the education budget discussed above. These findings suggest that the federal stimulus funds To analyze how New Jersey school finances were affected during tempered the negative effect of the recession, at least in these the Great Recession and the ARRA federal stimulus period, we categories. In contrast, other noninstructional categories such conduct a trend-shift analysis using the following specification: as transportation and utilities suffer statistically significant declines from trend during this period. Conversations with New Jersey Department of Education staff revealed that the Yit = α1T + α2v1 + α3v2+ α3Xit + fi + εit, state, faced by constrained budget conditions, cut back on nonessential transportation costs, such as courtesy busing.14 where Yit represents a school finance variable for school district i in year t; T represents the time trend and takes a value of 0 in the We find this information is consistent with the above patterns immediate pre-recession year (2008) and increases in increments in transportation spending evident in our data. of 1 for each subsequent year and decreases by 1 in each previous Patterns suggest that New Jersey tried to maintain conti- nuity in the expenditure categories most related to student year; fi denotes school district fixed effects and controls for any learning and development. Instructional expenditure, which fixed characteristics of districts;X it denotes controls for racial composition and poverty level (percentage of students eligible includes teacher salaries and classroom expenses, constitutes the spending category that most directly supports students’ for free and reduced price lunches) of the district; v1 =1 if year ≥ a learning. With ARRA funds coming in, there is no evidence 2009 and 0 otherwise; v 2 =1 if year ≥ 2010 and 0 otherwise. of the negative effects on this category seen in the year prior The coefficientα 1 represents the overall trend in the corre- sponding financial variable during the pre-recession period. The to the stimulus. Like instructional expenditure, instructional support, student services, and student activities closely relate coefficients of interest are α2, representing the intercept shift at to the development of the student. These categories, combined the onset of the recession, and α 3 representing the additional in- tercept shift during the federal stimulus period. In Tables 1-6, we with instructional expenditure, are arguably the categories that most directly impact a students’ access to a “thorough and define α2 as “Recession” and α3 as “Stimulus.” The shifts relative efficient” education. In summary, our results show that the to preexisting trends in 2009 and 2010 are captured by α 2 and post-recession period was characterized by a shift in composi- (α2 + α 3), respectively. tion of expenditures in favor of categories that are linked most a Local, state, and federal governments finalize their budgets in the spring prior to the budgeted year. More specifically, the budgets for the 2008 closely to students’ learning and development. school year were finalized in the spring of 2007, before the recession Columns 7 and 8 in Table 3, panel B, investigate the Great officially began (in December 2007, as defined by the National Bureau Recession’s impact on median teacher and administrator sala- of Economic Research), and before decision makers were aware of the ries. Teacher salaries show an upward shift in both years after impending recession. Therefore, 2008 is considered pre-recession in our analysis of financial variables, and 2009 is taken as the first year budgets the recession; administrator salaries show a downward shift in were directly affected by recession. the first year, which turns positive in the second year. To understand and interpret these results, there are two key factors to consider. First, education personnel retirements spiked during this period, as rumors of potential pension funding cuts spread across districts.15 Recall that New Jersey aid to New Jersey. Map 2 provides additional illustration of the is one of the few states in which the state funds pensions, and increased reliance on federal aid in 2008 (the immediate pre-re- with state revenue streams depleted, pensions were seen as a cession year) and 2010. It shows a nearly across-the-board probable area to cut. Since teachers and administrators at the increase in the federal share of total aid in 2010. Finally, Table 1, age of retirement tend to have the highest salaries, an increase panel B, column 10, shows evidence in favor of negative shifts in retirements would logically lead to a decline in the overall in student enrollment in both years after the recession. median salary. This result is not what we see for teacher salary, Next, we analyze whether the various expenditure catego- although the increase in administrative retirements in 2009 is ries follow the declining path experienced by aggregate expen- consistent with the negative shift seen in median administra- diture in the years after the Great Recession hit. Table 2 and Chart 6 present this analysis. Interestingly, there is marked 14 Courtesy busing is the nonmandatory provision of busing such as for variation. While instructional expenditure suffers a negative students living within walking distance of the school or who otherwise have a shift from trend in 2009, there is no evidence of any negative reasonable alternative to busing. effect in 2010. This pattern is repeated for instructional sup- 15 Surmised using data available from the State of New Jersey Department of port per pupil and student services per pupil. The resilience the Treasury’s Division of Pensions and Benefits.
50 Precarious Slopes C Patterns in Revenue during the Financial Crisis and Federal Stimulus Period
Percentage 25 Shift in 2009-10 20 * 15 Shift in 2008-09 10 * 5 * * * 0 * * * * -5 * * * -10 * * * * -15 * -20 * Total Total Federal Aid State Aid Local Aid Property Contribution Contribution Contribution Expenditures Funding per Pupil per Pupil per Pupil Taxes from from from per Pupil per Pupil per Pupil Federal Aid State Aid Local Aid
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary and Taxpayers’ Guide to Education Spending. Note: e * symbol denotes signicance at the 10, 5, or 1 percent level.
C There is more to the story, however. To understand fully Trends in State Aid to Teachers’ Pension Funds the patterns in teacher and administrator salaries, we consider Dollars per pupil Dollars per pupil a second factor: nontenured dismissal. In New Jersey, public 8,000 Total state aid 6,500 school employees attain tenure in their third year of employ- Scale 7,500 5,500 ment. With tenure, it becomes very difficult for an employee to be fired without extraordinary cause.16 As a result, the 7,000 4,500 State aid less vast majority of layoffs in New Jersey public education affect 6,500 TPAF funding 3,500 employees in their first and second years, reducing the num- Scale ber of employees at lower-level salaries. As Table 3, column 6,000 2,500 1, shows, there is strong evidence of large positive shifts in TPAF funding 5,500 Scale 1,500 teachers’ years of experience in both 2009 and 2010 relative to 5,000 500 trend, and both these effects are highly statistically significant. 1999 00 01 02 03 04 05 06 07 08 09 10 These results support the hypothesis that dramatic cuts in the number of lower-level employees increase the overall medi- Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary and Taxpayers’ Guide to Education an teacher salary significantly in both years. Administrative Spending. employees’ years of experience also showed a positive shift Notes: Years denote spring terms. TPAF is Teacher’s Pension in 2010, although this finding is not statistically significant. Annuity Fund. These patterns provide evidence that the significant positive shifts in median teacher and administrative salaries are likely due to a culling of lower-level public education employees during the post-recession era. tive salary. This relationship is corroborated by the patterns we observe in median years of experience of administrators. Though not statistically significant, there is a small decline in administrators’ years of experience in 2009 revealed in Table 3, column 2, a factor potentially contributing to the decline in administrators’ median salary. 16 New Jersey Statutes, Section 18a:6-10.
FRBNY Economic Policy Review / December 2013 51 M Percentage of District Revenue from Federal Sources
2007-08 2009-10
2008 overall average: 2010 overall average: 2.4 percent 3.3 percent
0 - 2 percent 3 - 4 percent 5 - 6 percent 7 - 8 percent 9 - 24 percent
Sources: Authors’ calculations, using the U.S. Census Bureau shapeles and the New Jersey Department of Education’s Audit Summary and Taxpayers’ Guide to Education Spending.
5.2 Examining Heterogeneities by School Discussions about spending on education in New Jersey are District Poverty Level most often framed in reference to wealth levels. In this vein, Table 4 and Chart 7 present the results for variations by pov- 17 While the above analysis focuses on aggregate patterns, the erty level. Affluent districts fared best in terms of preserving rest of the paper investigates whether there were differences instructional expenditure as well as most of noninstructional in impacts within the state by various characteristics, such expenditure (instructional support, student services, and as poverty status, location, and urbanicity. In the interest of transportation). They also experience the largest upward space, we focus here only on a subset of the finance indicators shifts in median teacher salaries and years of experience in of most interest: components of expenditure. This analysis both years after the recession. The combined results for salary provides useful insight into how the different districts allocat- and experience imply that affluent districts may have had the ed funds, and how the students in these districts were affected. largest instance of lower-level teacher layoffs. Affluent districts Results for the other indicators are available on request. 17 Charts 7-10 are placed in the “Conclusion” of this article (pages 16-24).
52 Precarious Slopes? Table 2 Patterns in the Composition of Expenditures during the Financial Crisis and Fiscal Stimulus Period
(1) (2) (3) (4) Instructional Expenditures Instructional Support Student Services Transportation Panel A per Pupil per Pupil per Pupil per Pupil
Percentage shift in 2008-09 -2.14*** -2.12** -2.0** -3.21*** Percentage shift in 2009-10 0.14 -0.86 0.97 -5.41***
Pre-recession base 7,787 1,909 1,599 763
Trend 165.4*** 66.3*** 57.4*** 17.2*** (5.9) (2.3) (1.8) (1.1) Recession -166.5*** -40.5** -31.9** -24.5*** (38.9) (18.2) (14.4) (9.1) Stimulus 177.6*** 24.2 47.4** -16.8 (48.9) (25.0) (19.2) (12.3)
Observations 6,752 6,752 6,752 6,744 R2 0.627 0.704 0.742 0.825
(5) (6) (7) (8) Student Activities Utilities and Maintenance Median Teacher Median Administrative Panel B per Pupil per Pupil Salary Salary
Percentage shift in 2008-09 0.81 -2.35*** 1.32*** -1.16*** Percentage shift in 2009-10 1.18 -4.50*** 6.45*** 1.55***
Pre-recession base 238 1,615 57,598 107,074
Trend 5.0*** 49.3*** -387.8*** 130.9* (0.3) (1.6) (39.2) (73.0) Recession 1.9 -37.9*** 761.3*** -1,239.4*** (2.5) (11.8) (215.8) (457.4) Stimulus 0.9 -34.9** 2,956.5*** 2,896.3*** (3.1) (15.0) (253.7) (592.5)
Observations 6,685 6,752 5,614 5,605 R2 0.959 0.728 0.815 0.787
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data.
Notes: Robust standard errors are in parentheses. All regressions control for racial composition and percentage of students eligible for free or reduced-price lunch, and include school district fixed effects. Pre-recession base is expressed in 2010 constant dollars.
***Statistically significant at the 1 percent level. **Statistically significant at the 5 percent level. *Statistically significant at the 10 percent level.
have the smallest declines in expenditures on utilities in 2009, 2010; while high-poverty districts show large, statistically sig- but their experience in 2010 is not very different from that of nificant declines, the affluent districts show large, statistically high- and medium-poverty districts in this category. significant increases. The variables for student services and -in The most noteworthy pattern revealed in this analysis is the structional support capture expenditures on services that are comparatively large declines in spending in both instructional designed to support, assess, and improve students’ well-being. and noninstructional categories. The most disparate examples They include social work, health services, technology, library are the shifts in student services and instructional support in costs, and student guidance.
FRBNY Economic Policy Review / December 2013 53 C Patterns in Expenditures during the Financial Crisis and Federal Stimulus Period
Percentage 6 Shift in 2008-09 * 4 Shift in 2009-10
2
0 * * * -2 *** * * -4 * -6 Instructional Instructional Student Transportation Student Utilities and Median Teacher Median Expenditures Support Services per Pupil Activities Maintenance Salary Administrative per Pupil per Pupil per Pupil per Pupil per Pupil Salary
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data. Note: e * symbol denotes signi cance at the 10, 5, or 1 percent level.
Table 3 5.3. Examining Heterogeneities by Patterns in Years of Experience during the Financial Crisis and Fiscal Stimulus Period Urbanicity
(1) (2) Another characteristic of school districts frequently cov- Median Teacher Median Administrator ered in discussions about New Jersey education financing is Years of Experience Years of Experience urbanicity. Historically, since urban districts generally have Percentage shift in 2008-09 8.80*** -0.10 lower property values and more apartment buildings housing Percentage shift in 2009-10 15.96*** 1.91 multiple families, the ratio of students to potential sources of Pre-recession base 10.13 20.57 tax income is higher than for suburban or rural districts. The result has been a large disparity between the per-pupil aid Trend -0.42*** -0.45*** available for wealthier, rural districts compared with poorer, Recession 0.89*** -0.02 urban districts. Stimulus 0.73*** 0.41 Table 5 and Chart 8 analyze variations by urban, suburban, Observations 5,614 5,605 and rural status.18 Once again, while all three groups exhibit R2 0.72 0.593 statistically significant declines in instructional expenditure in Source: Authors’ calculations, using the New Jersey Department of Educa- 2009, there is no evidence of negative effects in 2010, sug- tion’s Report Card data. gesting stimulus funding helped offset the trend. This pattern Notes: Robust standard errors are in parentheses. All regressions control repeats consistently throughout our results, suggesting that for racial composition and percentage of students eligible for free or re- the districts in general strive to preserve instructional expen- duced-price lunch, and include school district fixed effects. Pre-recession diture. While the decline in instructional expenditure in 2009 base is expressed in 2010 constant dollars. is most pronounced for urban districts, the experiences in ***Statistically significant at the 1 percent level. 2010 are very similar across the three groups. **Statistically significant at the 5 percent level. In most noninstructional categories (instructional support, *Statistically significant at the 10 percent level. student services, transportation, and student activities), the rural districts fare the best, while the urban districts fare the worst. The experiences of the three groups are very similar for expenditures on utilities.
18 Tables 5 and 6 are placed in the “Conclusion” of this article (pages 16-24).
54 Precarious Slopes? Table 4 Heterogeneities by School District Poverty Level
Instructional Expenditures per Pupil Instructional Support per Pupil Student Services per Pupil
High Medium High Medium High Medium Panel A Poverty Poverty Affluent Poverty Poverty Affluent Poverty Poverty Affluent
Percentage shift in 2008-09 -3.19*** -2.78*** -1.24 -5.23** -2.15* -0.83 -4.63** -2.69** -0.41 Percentage shift in 2009-10 -1.93 0.21 0.99 -8.06*** 0.94 3.20 -5.21** 1.34 4.61**
Pre-recession base 8,039 7,667 7,749 2,121 1,790 1,913 1,736 1,515 1,615
Trend 227.8*** 175.3*** 100.0*** 101.6*** 56.6*** 53.2*** 88.6*** 50.2*** 44.6*** (15.2) (8.0) (9.0) (7.6) (2.7) (3.3) (5.1) (2.3) (3.1) Recession -256.3*** -212.9*** -96.2 -111.0** -38.5* -15.9 -80.3** -40.8** -6.6 (92.2) (49.3) (64.5) (46.3) (20.2) (30.8) (35.0) (17.6) (27.1) Stimulus 101.2 229.0*** 173.1* -59.9 55.3** 77.1 -10.2 61.1** 81.1** (98.3) (70.1) (96.2) (58.8) (27.8) (49.1) (36.9) (26.5) (41.0)
Observations 1,682 3,240 1,828 1,682 3,240 1,828 1,682 3,240 1,828 R2 0.442 0.816 0.827 0.633 0.806 0.803 0.692 0.809 0.796
Transportation per Pupil Student Activities per Pupil Utilities and Maintenance per Pupil High Medium High Medium High Medium Panel B Poverty Poverty Affluent Poverty Poverty Affluent Poverty Poverty Affluent
Percentage shift in 2008-09 -5.83** -3.68*** -0.85 0.73 0.54 0.68 -2.20 -3.65*** -1.49 Percentage shift in 2009-10 -9.71*** -3.54 -4.57 2.91 -0.21 -0.14 -5.26*** -4.39*** -4.53***
Pre-recession base 729 774 780 193 241 281 1,721 1,581 1,568
Trend 28.1*** 17.4*** 6.1*** 4.7*** 5.9*** 3.9*** 64.3*** 51.2*** 35.1*** (2.7) (1.4) (2.3) (0.5) (0.4) (0.6) (4.0) (2.3) (2.9) Recession -42.5** -28.5*** -6.6 1.4 1.3 1.9 -37.8 -57.7*** -23.4 (17.8) (9.1) (26.0) (4.4) (3.5) (5.4) (25.6) (15.7) (21.3) Stimulus -28.3 1.1 -29.0 4.2 -1.8 -2.3 -52.7* -11.7 -47.6 (19.0) (17.2) (32.2) (5.0) (4.5) (6.7) (27.0) (23.4) (29.4)
Observations 1,682 3,239 1,821 1,657 3,211 1,815 1,682 3,240 1,828 R2 0.816 0.901 0.755 0.933 0.966 0.965 0.666 0.810 0.802
All three groups show positive shifts in median teacher and Camden. Note that there is substantial diversity with- salary and experience. However, rural districts show smaller in these areas by poverty and urbanicity. This fact makes spikes in both measures compared with the suburban and studying heterogeneities by poverty and urbanicity along with urban districts, suggesting that lower-level teacher layoffs may distinctions by MD all the more relevant. have been less prevalent in rural districts. Recall that Map 1 defines the metropolitan areas of New Jersey. Newark constitutes the northwest portion of the state and includes the most affluent districts. East of Newark, the Wayne district is second in terms of wealth and has the 5.4. Examining Heterogeneities by largest population of Hispanic and Asian students. The Edison Metropolitan Area districts are similar in demographics to Wayne, however, on average, Edison hosts larger districts. Camden districts have We next look at the variations by metropolitan area, analyzing the highest instance of poverty, the largest black student popu- New Jersey’s four largest metropolitan divisions: New York- lation, and the largest number of small-sized districts. White Plains-Wayne, Edison-New Brunswick, Newark-Union,
FRBNY Economic Policy Review / December 2013 55 Table 4 (continued) Heterogeneities by School District Poverty Level
Median Teacher Salary Median Teacher Years of Experience
High Medium High Medium Panel C Poverty Poverty Affluent Poverty Poverty Affluent
Percentage shift in 2008-09 0.77 1.30** 2.37*** 7.78*** 8.53*** 13.15*** Percentage shift in 2009-10 5.19*** 6.54*** 8.15*** 14.59*** 14.22*** 26.31***
Pre-recession base 57,492 57,312 58,286 10.28 10.55 9.12
Trend -54.7 -418.4*** -586.0*** -0.3*** -0.4*** -0.5*** (89.1) (57.7) (70.7) (0.0) (0.0) (0.0) Recession 443.5 746.8** 1,379.9*** 0.8*** 0.9*** 1.2*** (434.5) (313.0) (415.5) (0.2) (0.2) (0.2) Stimulus 2,539.8*** 3,000.4*** 3,372.1*** 0.7*** 0.6*** 1.2*** (431.5) (384.7) (510.1) (0.2) (0.2) (0.3)
Observations 1,397 2,711 1,506 1,397 2,711 1,506 R2 0.821 0.826 0.834 0.750 0.744 0.663
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data.
Notes: Robust standard errors are in parentheses. All regressions control for racial composition and percentage of students eligible for free or reduced-price lunch, and include school district fixed effects. Pre-recession base is expressed in 2010 constant dollars.
***Statistically significant at the 1 percent level. **Statistically significant at the 5 percent level. *Statistically significant at the 10 percent level.
Table 6 and Chart 9 show our findings. While all four MDs and expenditure, relative to trend, following the recession in New suffer declines in instructional expenditure in 2009, these Jersey. Federal stimulus sparks improvements in 2010; while both patterns reverse in 2010 when they each shift upward slightly. variables still exhibit declines, they are somewhat smaller than in Camden endures the largest decline in 2009, while the MDs for 2009. There is also strong evidence of substitution of funds on the the most part see similar positive reversals in 2010. The excep- aid side. The infusion of funds from the federal stimulus occurs tion is Edison, which experiences an upward shift about double simultaneously with statistically and economically significant the size of the other MDs. In most of the noninstructional cate- cuts in state and local financing, especially the former. As a result, gories, Edison stands out as having the largest upward shifts. All relative reliance on federal aid increases in 2010, while reliance MDs show positive shifts in median teacher salaries and years on state aid declines. Without the support of the federal stimulus of experience, with Wayne showing the largest increase in both in 2010, our results suggest that total aid to districts would have years in both categories. In summary, our results show quite a declined significantly more. lot of variation across MDs, providing evidence that New Jersey Our results also show that the post-recession period is regions reacted differently to a lack of resources. characterized by a compositional shift in expenditures in favor of categories linked most closely to student learning. The categories for instructional expenditure, instructional support, student services, and student activities are preserved in 2010 6. Conclusion when districts receive ARRA support.19 In contrast, trans- portation and utilities expenditures decline, suggesting that This paper explores how school finances in New Jersey were -af fected during the Great Recession and the ARRA federal stimulus 19 While instructional support saw a small negative shift of less than 1 percent funding that followed. The analysis yields some interesting results. in 2010, it is statistically not different from zero and considerably smaller than the negative shift in 2009, suggesting stimulus funding helped to moderate the There is strong evidence of a downward shift in both total funding recession’s negative effects and preserve funding approximately at trend level.
56 Precarious Slopes? Chart 7 Heterogeneities by School District Poverty Level Shift in 2008-09 Shift in 2009-10 Percent Percent Percent 1 5 6 Instructional Expenditures per Pupil Instructional Support per Pupil Student Services per Pupil 4 0 * 0 2 -1 * 0 -2 * -5 * -2 * -3 * * -4 * * -4 -10 -6 High Medium Af uent High Medium Af uent High Medium Af uent Poverty Poverty Poverty Poverty Poverty Poverty
2 Transportation per Pupil 4 Student Activities per Pupil 1 Utilities and Maintenance per Pupil 0 0 3 -1 -2 2 -2 -4 * 1 -3 -6 * * -4 * 0 * -8 -5 * -10 * -1 -6 High Medium Af uent High Medium Af uent High Medium Af uent Poverty Poverty Poverty Poverty Poverty Poverty 10 Median Teacher Salary 30 Median Teacher Years of Experience 8 25 * * 20 6 * 15 * 4 * * 10 * 2 * * 5 * 0 * 0 High Medium Af uent High Medium Af uent Poverty Poverty Poverty Poverty
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data. Note: The * symbol denotes significance at the 10, 5, or 1 percent level.
policymakers prioritized spending on categories most related In addition to studying the overall impact of the recession, to student learning and development. we examine whether its effects varied by poverty, urban status, We also find some interesting patterns in the teacher and and metropolitan area. We find considerable heterogeneity. administrator labor market. The shifts in median salary and For example, the high-poverty and urban groups sustain the years of experience suggest a culling of lower-level public largest declines (relative to their respective trends) in the education employees during the post-recession era, perhaps post-recession era. The most extreme examples in the pover- driven by New Jersey’s tenure rules, which make it difficult to ty-level heterogeneities are the shifts in spending on student lay off more experienced employees. services and instructional support in 2010. The high-poverty
FRBNY Economic Policy Review / December 2013 57 Table 5 Heterogeneities by School District Urbanicity
Instructional Expenditures per Pupil Instructional Support per Pupil Student Services per Pupil
Panel A Urban Suburban Rural Urban Suburban Rural Urban Suburban Rural
Percentage shift in 2008-09 -3.83*** -1.89*** -2.24** -4.58 -2.3** 0.16*** -3.30 -2.00* -0.98 Percentage shift in 2009-10 0.80 0.21 0.02 -2.88 -1.35 2.21 -1.18 0.94 2.00
Pre-recession base 8,210 7,820 7,511 2,181 1,954 1,640 1,832 1,641 1,351
Trend 245.1*** 154.5*** 183.6*** 92.1*** 69.7*** 53.5*** 85.7*** 59.5*** 47.0*** (19.7) (7.9) (9.5) (8.8) (3.1) (3.7) (8.0) (2.3) (3.0) Recession -314.5** -147.5*** -168.1** -99.8 -44.9** 2.7 -60.5 -32.8* -13.3 (152.4) (47.2) (67.3) (85.6) (21.7) (32.2) (68.1) (17.0) (26.9) Stimulus 380.5* 164.1*** 169.9* 37.1 18.6 33.6 38.9 48.3** 40.3 (206.8) (58.2) (86.8) (108.3) (30.3) (40.3) (87.1) (22.7) (35.3)
Observations 440 5,041 1,271 440 5,041 1,271 440 5,041 1,271 R2 0.797 0.591 0.821 0.828 0.679 0.756 0.84 0.719 0.774
Transportation per Pupil Student Activities per Pupil Utilities and Maintenance per Pupil
Panel B Urban Suburban Rural Urban Suburban Rural Urban Suburban Rural
Percentage shift in 2008-09 -6.16 -4.48*** 1.09 0.52 0.47 2.96 -1.70 -2.28*** -2.47 Percentage shift in 2009-10 -11.78*** -5.85*** -2.46 0.42 0.63 4.86 -4.51*** -4.35*** -4.71**
Pre-recession base 691 701 1,037 181 259 172 1,743 1,631 1,512
Trend 21.0*** 14.4*** 21.8*** 5.6*** 5.7*** 2.7*** 78.5*** 45.8*** 52.9*** (3.1) (1.3) (2.4) (0.8) (0.4) (0.6) (4.6) (2.1) (2.7) Recession -42.6 -31.4*** 11.3 0.9 1.2 5.1 -29.7 -37.2*** -37.4 (26.7) (10.8) (19.2) (6.5) (2.9) (5.8) (47.3) (13.9) (23.1) Stimulus -38.9 -9.6 -36.8 -0.2 0.4 3.3 -48.9 -33.8* -33.8 (30.0) (14.9) (25.1) (9.1) (3.7) (7.1) (52.9) (17.5) (31.7)
Observations 440 5,033 1,271 432 5,009 1,244 440 5,041 1,271 R2 0.905 0.802 0.847 0.968 0.954 0.971 0.898 0.701 0.787
districts show large, statistically significant declines in these components of expenditures. In fact, some of this pressure is categories, while the affluent districts show large, statistically already evident. significant increases. These variables capture the expenditure Using a compilation of the annual budgets for the United on services to support, assess, and improve students’ well-be- States and the state of New Jersey, we plot budgeted and actual ing, including social work, health services, technology, library funding per pupil over 2000-12 in Chart 10. The chart shows costs, and student guidance. a noticeable decline in budgeted funding after 2010. It also Since New Jersey spent its appropriation of ARRA funds in reveals that New Jersey planned for steeper declines in 2011 2010, a valid question here is how we might expect the state compared with the nation as a whole. to fare in the near future. Considering the slow recovery of The state’s budget for the 2011 school year explicitly states economic activity and employment, state and local revenues that funds were not available to replace the ARRA funding of will likely continue to come in below trend. The end of the 2010. The funding levels required by the state’s SFRA formula federal stimulus funding and lower-than-trend growth in state were not met, and the 2011 budget shows statewide declines, and local revenues could lead to more significant downward with many districts’ aid being reduced as much as 5 percent pressure on funding and expenditures, including the various from the previous year. Cuts were made in many expendi-
58 Precarious Slopes? Table 5 (continued) Heterogeneities by School District Urbanicity
Median Teacher Salary Median Teacher Years of Experience
Panel C Urban Suburban Rural Urban Suburban Rural
Percentage shift in 2008-09 3.42** 1.51*** 0.07 9.17** 10.32*** 4.29** Percentage shift in 2009-10 8.05*** 7.13*** 3.33*** 17.43*** 18.58*** 7.72***
Pre-recession base 58,634 57,838 56,291 10.90 9.69 11.66
Trend -446.8** -476.8*** -108.3* -0.4*** -0.5*** -0.3*** (179.7) (48.8) (65.3) (0.1) 0.0 0.0 Recession 2,005.5** 873.6*** 40.6 1.0** 1.0*** 0.5** (823.6) (253.6) (447.6) (0.4) (0.1) (0.3) Stimulus 2,712.8*** 3,250.6*** 1,834.1*** 0.9* 0.8*** 0.4 (987.9) (296.5) (547.8) (0.5) (0.1) (0.3)
Observations 366 4,192 1,056 366 4,192 1,056 R2 0.882 0.795 0.853 0.728 0.699 0.741
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data.
Notes: Robust standard errors are in parentheses. All regressions control for racial composition and percentage of students eligible for free or reduced-price lunch, and include school district fixed effects. Pre-recession base is expressed in 2010 constant dollars.
***Statistically significant at the 1 percent level. **Statistically significant at the 5 percent level. *Statistically significant at the 10 percent level.
ture categories, leaving the planned expansion of a preschool The authors thank Jason Bram, Erica Groshen, Andrew Haughwout, program stalled, special education allocations and nonpublic James Orr, Joydeep Roy, Amy Ellen Schwartz, Leanna Stiefel, Giorgio school aid 15 percent below projected need, debt service aid Topa, and seminar participants at the Federal Reserve Bank of New down by 15 percent, and funding for adult education slashed York and the New York University-FRBNY Fiscal Breakfast for valu- entirely. Data from the state’s Department of Education show able insight and feedback. They are grateful to Kevin Dehmer, Susan that in 2011 the number of full- and part-time public school Ecks, Frank Lavdas, and the New Jersey Department of Education for teachers in New Jersey dropped 4 percent, while the number patiently answering numerous questions and providing generous help of administrators fell 7 percent. in acquiring the data. They also thank staff at the U.S. Census Bureau, As economists are predicting continued softness, school the U.S. Department of Education, and the New Jersey School Boards districts will likely face hard decisions ahead involving cuts Association for answering many questions and providing assistance to the critical instructional expenditure category that they with different parts of the data. Elizabeth Setren provided excellent have so far been successful in preserving. This possibility research assistance. could have adverse effects on human capital formation and, by extension, the nation’s future. Our findings form an important basis for understanding schools’ financial situations during recessions and can serve to guide future policy decisions.
FRBNY Economic Policy Review / December 2013 59 Chart 8 Heterogeneities by School District Urbanicity Shift in 2008-09 Shift in 2009-10 Percent Percent Percent 2 Instructional Expenditures per Pupil 4 Instructional Support per Pupil 4 Student Services per Pupil 1 2 2 0 0 -1 0 * -2 * -2 * -2 * -3 -4 -4 * -6 -4 Urban Suburban Rural Urban Suburban Rural Urban Suburban Rural
5 Transportation per Pupil 6 Student Activities per Pupil 1 Utilities and Maintenance per Pupil 5 0 0 4 -1 * -5 * 3 -2 * 2 -3 -10 * 1 -4 * * * -15 -5 Urban Suburban Rural Urban Suburban Rural Urban Suburban Rural
10 Median Teacher Salary 30 Median Teacher Years of Experience 8 25 * 20 6 * 15 * * 4 10 * * * 2 * 5 * * * 0 0 Urban Suburban Rural Urban Suburban Rural
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data. Note: The * symbol denotes significance at the 10, 5, or 1 percent level.
60 Precarious Slopes? Table 6 Heterogeneities by School District Metropolitan Area
Instructional Expenditures per Pupil Instructional Support per Pupil Student Services Per Pupil
Panel A Camden Edison Newark Wayne Camden Edison Newark Wayne Camden Edison Newark Wayne
Percentage shift -2.58*** -1.56 -2.46*** -1.22 -1.99 0.90 -1.48 -3.87 -3.17 1.39 -1.26 -3.38 in 2008-09 Percentage shift 0.43 1.30 0.24 0.16 -1.85 4.30 0.89 -4.95 -2.10 5.55* 3.09* -1.20 in 2009-10
Pre-recession base 7,314 7,783 7,974 7,862 1,757 1,868 1,998 2,031 1,464 1,594 1,692 1,676
Trend 176.2*** 185.7*** 146.4*** 79.1*** 68.5*** 68.5*** 67.1*** 59.3*** 58.7*** 63.1*** 56.7*** 53.8*** (11.1) (18.3) (8.8) (9.4) (4.4) (6.2) (3.6) (7.2) (3.5) (5.0) (3.1) (4.8) Recession -189.0*** -121.8 -196.2*** -96.1 -35 16.9 -29.5 -78.6* -46.4 22.2 -21.3 -56.6 (67.7) (104.2) (65.6) (70.1) (36.2) (39.8) (33.4) (44.0) (28.4) (36.2) (27.8) (36.2) Stimulus 220.6** 223.0* 214.9** 108.9 2.6 63.4 47.4 -22.1 15.7 66.3 73.5** 36.5 (88.2) (128.8) (84.6) (82.8) (41.9) (56.4) (42.1) (72.1) (32.9) (49.5) (36.9) (42.8)
Observations 1,252 1,424 1,605 1,260 1,252 1,424 1,605 1,260 1,252 1,424 1,605 1,260 R2 0.784 0.368 0.783 0.842 0.752 0.518 0.776 0.8 0.768 0.577 0.79 0.836
Transportation per Pupil Student Activities per Pupil Utilities and Maintenance per Pupil
Panel B Camden Edison Newark Wayne Camden Edison Newark Wayne Camden Edison Newark Wayne
Percentage shift -6.42*** -4.35** -1.06 -2.83 1.38 -0.60 -0.23 2.72 -2.16 -0.17 -2.69** -2.41 in 2008-09 Percentage shift -9.2*** -7.47*** -4.27** -3.95 0.75 1.56 0.76 0.16 -4.50*** -2.22 -5.48*** -4.86*** in 2009-10
Pre-recession base 726 811 822 612 223 253 264 269 1,540 1,659 1,616 1,616
Trend 23.8*** 15.5*** 11.8*** 4.2 4.6*** 6.2*** 4.5*** 5.1*** 49.9*** 55.6*** 43.3*** 34.0*** (2.3) (2.2) (2.0) (3.3) (0.7) (0.7) (0.7) (0.7) (3.2) (4.7) (2.6) (3.2) Recession -46.6*** -35.3** -8.8 -17.4 3.1 -1.5 -0.6 7.3 -33.2 -2.9 -43.5** -39.0* (15.0) (17.8) (13.6) (30.4) (5.3) (5.7) (5.9) (5.2) (22.3) (29.2) (21.4) (23.1) Stimulus -20.1 -25.3 -26.4 -6.8 -1.4 5.5 2.6 -6.9 -36.1 -33.9 -45.1* -39.5 (18.8) (21.1) (16.8) (35.4) (6.6) (7.5) (6.9) (6.1) (27.3) (37.1) (26.8) (30.4)
Observations 1,252 1,424 1,597 1,260 1,229 1,412 1,595 1,259 1,252 1,424 1,605 1,260 R2 0.897 0.777 0.916 0.738 0.959 0.946 0.962 0.971 0.849 0.482 0.813 0.899
FRBNY Economic Policy Review / December 2013 61 Table 6 (continued) Heterogeneities by School District Metropolitan Area
Median Teacher Salary Median Teacher Years of Experience
Panel C Camden Edison Newark Wayne Camden Edison Newark Wayne
Percentage shift 0.02 1.87** 1.89*** 2.87*** 3.62 11.78*** 12.51*** 13.11*** in 2008-09 Percentage shift 4.27*** 7.64*** 7.76*** 9.47*** 7.24*** 22.48*** 21.9*** 25.14*** in 2009-10
Pre-recession base 56,138 55,400 58,511 60,527 11.05 9.34 9.59 9.15
Trend -66.3 -493.4*** -607.1*** -1,087.2*** -0.3*** -0.5*** -0.5*** -0.7*** (74.3) (79.1) (76.0) (105.8) 0.0 0.0 0.0 0.0 Recession 11.7 1,037.6** 1,106.8*** 1,734.9*** 0.4 1.1*** 1.2*** 1.2*** (436.7) (471.4) (427.6) (560.0) (0.3) (0.2) (0.2) (0.2) Stimulus 2,383.8*** 3,195.2*** 3,430.8*** 3,995.3*** 0.4 1.0*** 0.9*** 1.1*** (530.4) (559.4) (504.9) (617.4) (0.3) (0.3) (0.2) (0.3) 1,042.0 1,181.0 1,335.0 1,047.0 Observations 1,042 1,181 1,335 1,047 R2 0.809 0.788 0.813 0.812 0.718 0.692 0.694 0.698
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data.
Notes: Robust standard errors are in parentheses. All regressions control for racial composition and percentage of students eligible for free or reduced-price lunch, and include school district fixed effects. Pre-recession base is expressed in 2010 constant dollars.
***Statistically significant at the 1 percent level. **Statistically significant at the 5 percent level. *Statistically significant at the 10 percent level.
62 Precarious Slopes? Chart 9 Heterogeneities by School District Metropolitan Division Shift in 2008-09 Shift in 2009-10 Percent Percent Percent 6 8 2 Instructional Expenditures per Pupil Instructional Support per Pupil Student Services per Pupil 4 6 * 1 * 2 4 * 0 0 2 * -1 -2 0
-2 -4 * -2 * * -3 -6 -4
ayne ayne W ayne Edison W Camden Edison Newark Camden Edison Newark W Camden Newark 2 3 1 Transportation per Pupil Student Activities per Pupil Utilities and Maintenance per Pupil 0 0 2 -1 -2 -2 -4 * * 1 * -3 -6 * -4 * 0 * -8 * * -5 * -10 -1 -6
ayne ayne ayne Edison W Edison W Camden Edison Newark W Camden Newark Camden Newark 10 30 Median Teacher Salary Median Teacher Years of Experience * 25 8 * * * 20 * * 6 15 4 * 10 * * * 2 * 5 * ** 0 0
ayne W Edison ayne Camden Edison Newark Camden Newark W
Sources: Authors’ calculations, using the New Jersey Department of Education’s Audit Summary, Taxpayers’ Guide to Education Spending, and Report Card data. Note: e * symbol denotes signi cance at the 10, 5, or 1 percent level.
FRBNY Economic Policy Review / December 2013 63 C Total Funding per Pupil: Budgeted versus Actuall
In ation-adjusted dollars 20,000 New Jersey budgeted 18,000 New Jersey actual 16,000 U.S. budgeted 14,000 U.S. actual 12,000
10,000 2000 01 02 03 04 05 06 07 08 09 10 11 12
Source: Authors’ calculations, using U.S. Department of Education and New Jersey State Treasury data.
64 Precarious Slopes? References
Baker, B. 2009. “Within-District Resource Allocation and the Firestone, W. A., M. E. Goertz, B. Nagle, and M. F. Smelkinson. 1994. Marginal Costs of Providing Equal Educational Opportunity: “Where Did the $800 million Go? The First Year of New Jersey’s Evidence from Texas and Ohio.” Education Policy Analysis Quality Education Act.” Education Evaluation and Policy Archives 17, no. 3 (February):1-31. Analysis 16, no. 4 (winter): 359-73.
Barro, R. J. 1991. “Economic Growth in a Cross Section of Coun- Firestone, W. A., M. E. Goertz, and G. Natriello. 1997. From Cash- tries.” Quarterly Journal of Economics 106, no. 2 (May): box to Classroom: The Struggle for Fiscal Reform and 407-43. Educational Change in New Jersey. New York: Teachers College Press. Becker, G. S. 1994. Human Capital: A Theoretical and Empiri- cal Analysis with Special Reference to Education. Chica- Gerst, J., and D. Wilson. 2010. “Fiscal Crises of the States: Causes go: University of Chicago Press. and Consequences.” Federal Reserve Bank of San Francisco Economic Letter 2010-20, June 28. Bedard, K., and W. O. Brown, Jr. 2000. “The Allocation of Public School Expenditures.” Claremont Colleges Working Papers, no. Gordon, N. 2004. “Do Federal Grants Boost School Spending? Evi- 2000-16, August. dence from Title I.” Journal of Public Economics 88, no. 9-10 (August):1771-92. Benhabib, J., and M. M. Spiegel. 1994. “The Role of Human Capital in Economic Development: Evidence from Aggregate Cross-Coun- Hanushek, E. A., and L. Woessmann. 2007. “The Role of Education try Data.” Journal of Monetary Economics 34, no. 2 (Octo- Quality in Economic Growth.” World Bank Policy Research ber): 143-74. Working Papers, no. 4122, February.
Betts, J. R. 1995. “Which Types of Public School Spending Are Most Rubenstein, R., A .E. Schwartz, L. Stiefel, and H. B. H. Amor. 2007. Effective? New Evidence on the School Quality Debate.” Universi- “From Districts to Schools: The Distribution of Resources across ty of California at San Diego Department of Economics Discus- Schools in Big City School Districts.” Economics of Education sion Papers, no. 95-03. Review 26, no. 5 (October): 532-45.
Deitz, R., A. F. Haughwout, and C. Steindel. 2010. “The Recession’s Stiefel, L., and A. E. Schwartz. 2011. “Financing K-12 Education in Impact on the State Budgets of New York and New Jersey.” Federal the Bloomberg Years, 2002-2008.” In J. A. O’Day, C. S. Bitter, and Reserve Bank of New York Current Issues in Economics and L. M. Gomez, eds., Education Reform in New York City: Finance 16, no. 6 (June/July): 1-11. Ambitious Change in the Nation’s Most Complex School System. Cambridge: Harvard Education Press. Feldstein, M. 1978. “The Effect of a Differential Add-on Grant: Title I and Local Education Spending.” Journal of Human Resources 13, no. 4 (autumn): 443-58.
The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.
FRBNY Economic Policy Review / December 2013 65
Carlo Rosa
The Financial Market Effect of FOMC Minutes
• There is a vast array of evidence on the 1. Introduction financial market effect of monetary news released on Federal Open Market Committee any studies have examined the influence of the (FOMC) meeting days. MFederal Reserve’s unanticipated target rate decisions on U.S. asset prices.1 A recent strand of literature has also • Yet little is known about the real-time response looked at the asset price response to the Federal Open Market of U.S. asset prices to the information Committee (FOMC) statements.2 Despite the vast and contained in the FOMC minutes. growing empirical evidence on the financial market effect of monetary news released on FOMC meeting days, little is • This article uses a novel data set to examine known about the real-time response of U.S. asset prices to the effect of the FOMC minutes release on the information originating from central bank minutes. This U.S. asset prices. article fills the gap by using a novel, high-frequency data set to broaden the understanding. • The release is shown to significantly affect the Central bank communication has become increasingly volatility of U.S. asset prices and their trading transparent over the past decade. This is important not only volume, with the magnitude of the effects for reasons of democratic legitimacy and accountability but economically and statistically significant. also for monetary policy to be most effective (Woodford 2005). Central banks use many communication channels, • The asset price response to the FOMC including media statements, press conferences, speeches, minutes has declined since 2008, reports, and minutes. This article contrasts the effect of FOMC statement releases with that of FOMC minutes releases. These suggesting greater transparency two releases differ mainly in the amount and timeliness of by the Committee.
1 See, for example, Kuttner (2001), Rigobon and Sack (2004), Bernanke and Kuttner (2005), Fleming and Piazzesi (2005), Faust et al. (2007), and the references therein. 2 See, for instance, Gurkaynak, Sack, and Swanson (2005) and Rosa (2011a, 2011b).
Carlo Rosa is an economist in the Federal Reserve Bank of New York’s The author thanks Marcus Lee, William Riordan, Tony Rodrigues, and Markets Group. Andrea Tambalotti for useful comments as well as Michael Fleming and the [email protected] coeditor, Marco Del Negro, for detailed comments on an earlier draft that led to significant improvements. All remaining errors are the author’s. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System.
FRBNY Economic Policy Review / December 2013 67 their information. FOMC statements explain the rationale using trading volumes, redo the computations on a different for the policy action and convey the outlook for the future subsample, and perform a comparative exercise by looking monetary policy stance. FOMC minutes provide more at the financial market effect of the release of the Bank of detailed information on the range of Committee members’ England minutes. This sensitivity analysis corroborates the views on the appropriate policy stance, on the U.S. economic core finding that central bank minutes contain market- outlook, and on the near-term monetary policy inclination. relevant information, especially for fixed-income securities. The statement is released at the moment of the target rate The rest of the article is organized as follows. Section 2 decision, whereas the minutes come out three weeks after describes the data set. In section 3, I discuss the empirical the FOMC meets. The extent to which market participants results of the asset price reaction to the release of FOMC may scrutinize the FOMC minutes to gain information minutes. The robustness of the results is examined in beyond what is contained in the statement is a question to be section 4, followed by a conclusion in section 5. answered empirically. The article’s main findings can be summarized as follows. First, I examine the financial market effect of the release of FOMC minutes on U.S. asset prices (Treasury rates, stock prices, and 2. Data U.S. dollar exchange rates) using a high-frequency, event-study analysis. The use of intraday data allows for better isolation of The high-frequency data on U.S. asset prices I use consist the response of asset prices to the minutes release, since no other of quotes measured at five-minute intervals of on-the-run economic news is systematically released within such a narrow two- and ten-year Treasury yields, futures prices on the (five-minute) window around the monetary announcement. S&P 500 stock index, and the U.S. dollar exchange rate The release of the minutes is shown to induce “higher than against the euro, Swiss franc, and Japanese yen, covering normal” volatility across different asset classes. For instance, the the period January 2005 to March 2011. Prior to 2005, the volatility of two-year Treasury yields is roughly three times larger FOMC minutes were released only after the next meeting on event days than during a period free of such an event. This had finished, rendering them largely of historical interest. finding suggests that the FOMC minutes provide market-relevant The sample ends in March 2011 to exclude the period when information and is consistent with the results of Boukus and the FOMC started to release the Summary of Economic Rosenberg (2006) showing that the themes of the FOMC minutes Projections and to hold a press conference immediately are correlated with current and future economic conditions.3 after its meeting. Midpoints of bid/ask quotes or indicative Second, to gauge the importance of the minutes’ release, quotes, observed at the end of each five-minute interval, are I compare the increase in the variance of U.S. asset prices used to generate the series of (equally spaced) five-minute attributed to the minutes with the response brought about continuously compounded asset price returns.4 The Treasury by the release of the FOMC balance-of-risk statement, the bond yields are provided by Tradeweb and are based on nonfarm payroll macroeconomic announcement, and the indicative prices, rather than transaction prices.5 Hence, there Institute for Supply Management (ISM) manufacturing index are no associated volume data available. The S&P 500 futures (a purchasing survey of the U.S. manufacturing sector). The data refer to the E-Mini S&P, a stock market index futures financial market effect of the FOMC minutes is similar to that contract traded on the Chicago Mercantile Exchange’s Globex of the ISM manufacturing index, although smaller than the electronic trading platform, and consist of both prices and market effect induced by the FOMC statement and nonfarm trading volumes. A continuous series is constructed by payrolls, often referred to as the “king” of announcements by considering the front-month contract, and rolling over to market participants (Andersen and Bollerslev 1998). the next contract on expiration date. Foreign exchange data Third, I document that the asset price response to the are provided by EBS (Electronic Broking System, now part of minutes has declined in the recent period. One potential ICAP) and include trading volume in the global interdealer interpretation of this finding is that the statement has become more informative and that the FOMC has 4 For instance, Bandi and Russell (2008) argue that five-minute returns put more effort into greater transparency by releasing provide a reasonable balance between sampling too frequently (and confounding price reactions with market microstructure noise, such as the information in a timelier manner. bid-ask bounce, staleness, price discreteness, and the clustering of quotes) and Finally, the robustness of the above results is examined sampling too infrequently (and blurring price reactions to news). along several dimensions. For instance, I carry out the analysis 5 Although the use of market data may be preferred, the existing literature on exchange rates (for example, Phylaktis and Chen [2009] and Danielsson and Payne [2002]) has documented that indicative data bear no qualitative 3 Similarly, Apel and Blix Grimaldi (2012) show that the sentiment of the Sveriges difference from data on transaction quotes. Hence, it is extremely unlikely Riksbank minutes is useful in predicting the bank’s future policy rate decisions. that the results of this study are driven by the use of indicative quotes.
68 The Financial Market Effect of FOMC Minutes Summary Statistics
Swiss Franc/ Japanese Two-Year Treasury Ten-Year Treasury S&P 500 Euro/U.S. Dollar U.S. Dollar Yen/U.S. Dollar Mean 0 0 0 0 0 0 Median 0 0 0 0 0 0 Maximum 0.21 0.18 4.32 1.57 2.32 2.5 Minimum -0.44 -0.31 -2.95 -1.19 -2.27 -2.77 Standard deviation 0.01 0 0.09 0.04 0.05 0.05 Skewness -2.14 -2.63 0.61 0.12 0.35 0.08 Kurtosis 253 221 74 47 78 122 Jarque-Bera p-value 0 0 0 0 0 0 Observations 221,729 287,575 406,727 452,549 447,342 450,427
Source: Author’s calculations.
Notes: The table reports summary statistics for the variables used in the econometric analysis. The sample period is January 2005 to March 2011, excluding all weekend days. The asset price return is either the five-minute change in the bond yields or the five-minute percentage change in the stock price or the U.S. dollar exchange rate pairs.
spot market (see Chaboud et al. [2004] for a detailed content of the minutes is not always completely anticipated, description of the data).6As noted by Chaboud, Chernenko, the release of the minutes causes market participants to revise and Wright (2008), EBS and Reuters are two electronic their expectations, and this should be reflected in higher broking systems used globally for interdealer spot trading. volatility of asset prices compared with a period free of such Trading in the euro-dollar and dollar-yen currency pairs is an event. Since asset price volatility may be time-varying, concentrated primarily on EBS. it is important to properly control for both intraday and The table, which presents a selection of descriptive statistics day-of-the-week effects when gauging whether the release of for all variables used in this study, reveals that the mean and the minutes induces elevated price fluctuations. To that end, median of the five-minute bond yield changes and stock and Chart 1 displays 1) the standard deviation of the five-minute exchange rate returns are very close to zero. All returns are returns on release days and 2) the standard deviation of the approximately symmetric and all display excess kurtosis. The five-minute returns on the same weekdays (of the previous Jarque-Bera statistics strongly reject the null hypothesis that and following week of the release day of the FOMC minutes) returns are normally distributed. and hours, but on nonannouncement days.7 The vertical line is shown at the release time of the FOMC minutes, that is, 2 p.m. ET. The dark and white squares denote significance of the differences at the two-sided 1 and 5 percent levels, 3. Results respectively. Since asset price returns are not normally distributed, I use the test statistic proposed by Levene (1960) A model testing for the financial market effect of central to test the null hypothesis of equal variances in each subgroup. bank minutes would ideally identify the surprise component Fifty sets of FOMC minutes were published between January of their content. Unfortunately, there are no direct measures of market expectations about the information contained in the FOMC minutes. Hence, to get around the difficulties of 7 quantifying the surprise component, I follow the methodology More specifically, for both announcement______and nonannouncement days the T _ 2 standard deviation is defined as , where r is the of Kohn and Sack (2004) and look at whether, and to what √∑t=1 (rt - r) /(T -1) t five-minute return, T is the number of observations in the sample, andr_ is extent, the volatility of asset prices is higher on release days the sample mean. As a robustness check,______I also consider the squared root of T 2 compared with nonevent days. The idea is that as long as the the mean of squared returns, that is, r / ; the results (available √∑t=1 t (T -1) upon request) remain extremely similar. To compute “normal” U.S. asset price volatilities, that is, the volatility that would be expected to prevail on control 6 The foreign exchange trading volume data are proprietary; to preserve data (or nonevent) days, as a further robustness check I also use the previous and confidentiality, I report only relative volumes expressed in ratio form, rather following day of the release day of the FOMC minutes. It is reassuring that the than as actual amounts of base currency. results reported in Chart 1 continue to hold.
FRBNY Economic Policy Review / December 2013 69 C The Volatility of Asset Prices around FOMC Minutes Releases
1 percent signi cance 5 percent signi cance Standard deviation Standard deviation 0.020 0.014 Two-year Treasury yield Ten-year Treasury yield
0.012 FOMC minutes 0.015 release FOMC minutes 0.010 release
0.010 0.008
0.006 0.005 0.004
0.000 0.002
0.25 0.10 S&P 500 Euro/U.S. dollar
0.20 0.08 FOMC minutes FOMC minutes release release
0.15 0.06
0.10 0.04
0.02 0.05 0.10 0.10 Swiss franc/ Japanese yen/ U.S. dollar U.S. dollar
0.08 0.08 FOMC minutes FOMC minutes release release
0.06 0.06
0.04 0.04
0.02 0.02 12:55 14:00 15:00 12:55 14:00 15:00 Time Time
Source: Author’s calculations. Notes: e chart plots the standard deviation of ve-minute asset price returns around the FOMC minutes (solid line) and on control days (the same weekdays and hours of the previous and following weeks of the FOMC minutes release day; dashed line). e sample period is January 2005 to March 2011. e interval spans from one hour before to two hours a er the event time. e vertical line signi es the release time of the FOMC minutes, 2 p.m. ET. Levene (1960) statistics are employed to test the null hypothesis of equal variances in each subgroup. Dark and white squares denote signi cance of the di erences at the two-sided 1 and 5 percent level, respectively.
70 The Financial Market Effect of FOMC Minutes 2005 and March 2011.8 The release of the minutes induces FOMC statements and a much larger effect than does the significantly “higher than normal” volatility on asset prices, response of asset prices to the release of FOMC minutes.10 especially at the time of the release, and up to roughly one hour The response of ten-year Treasury rates and S&P 500 stock after the announcement. For instance, the volatility of two-year prices to nonfarm payrolls is smaller than the response Treasury yields suddenly jumps at the time of the release—it induced by the FOMC statement, whereas the U.S. dollar is roughly three times larger on event days compared with exchange rates are more sensitive to nonfarm payrolls than a period free of such an event, and it remains significantly to monetary news. To better assess the economic importance higher until around 3 p.m. ET. Treasuries, especially at shorter of the financial market effect of the release of the FOMC maturities, are the most affected asset class, closely followed by minutes, in panel C of Chart 2 I show that the release of the U.S. dollar exchange rates, whereas the response of stock prices ISM manufacturing index induces “higher-than-normal” is less pronounced, though still significantly higher than it is volatility that has roughly the same order of magnitude as the on nonevent days, and shorter-lived.9 “excess” volatility induced by the minutes.11 For instance, at This finding indicates that FOMC minutes provide the news release time (2 p.m. for the minutes and 10 a.m. for market-relevant information that is incorporated into asset the ISM manufacturing index), the volatility of the two-year prices. To gauge the order of magnitude of these effects, Treasury yield equals 0.016 (1.6 basis points) for both releases, I compare the increase in the volatility of U.S. asset prices compared with a “normal” volatility of 0.004. attributed to the minutes with that induced by the release of I also investigate whether the informational content of the the FOMC balance-of-risk statement, the nonfarm payroll FOMC minutes has changed over time by splitting the sample macroeconomic announcements (one of the most closely into two subsamples. Chart 3 displays 1) the standard deviation followed announcements by the financial press), and the of the five-minute returns on release days and 2) the standard ISM manufacturing index. Panel A of Chart 2 shows that the deviation of the five-minute returns on the same weekdays FOMC statement exerts an economically large and highly (of the previous and following week of the release day of the significant effect on asset prices. For instance, the ten-year minutes) and hours, but on nonannouncement days, for two rate, S&P 500 stock prices, and the euro-dollar exchange rate samples: January 2005-December 2007 in panel A and January are at least eight times more volatile on event days compared 2008-March 2011 in panel B. The chart documents that the with nonevent days. The least affected asset price is the overall level of volatility on nonevent days has increased during Japanese yen, but that is still four times as volatile as it is on the financial crisis, especially for stock prices. Moreover, the normal days. The absorption of news is also more prolonged, level of asset price volatility on release days has become more taking roughly one hour and thirty minutes, compared similar to the level of volatility on control days for 2008-11 with the time associated with the release of the FOMC compared with 2005-07. One potential interpretation of this minutes. As documented by Rosa (2011a), for U.S. stock finding is that FOMC communication before the release of and volatility indexes (the Dow Jones Industrial Average, the minutes has become more informative, possibly indicating NASDAQ 100, S&P 500, and VIX), after the initial effect the that the Committee has achieved greater transparency by market will seek its new equilibrium, taking into account the releasing news in a more timely manner. A complementary additional information generated by the stock price changes interpretation is that the sensitivity of asset prices and, in following the FOMC announcements and the subsequent particular, of interest rates, to news diminishes when short- commentaries on the Federal Reserve’s decisions provided term rates hit the zero lower bound. The evidence provided by in real time by financial analysts. Therefore, although the Swanson and Williams (2012), however, rejects this hypothesis. FOMC monetary news affects asset prices immediately, the market dynamics toward its new equilibrium are protracted and extend well beyond the initial effect. Consistent with the findings of Fleming and Remolona (1999) and Balduzzi, 10 The set of nonannouncement days for the nonfarm payroll release is Elton, and Green (2001), I show in panel B of Chart 2 that defined as follows. First, I run the Bloomberg function “ECO United nonfarm payrolls exert a similar effect on the release of States,” which provides time series data for all U.S. macroeconomic news stored by Bloomberg. Next, I select the same weekdays of the previous and following week of the release day of nonfarm payrolls. Finally, I define as 8 The release dates can be found on the Federal Reserve Board’s website, and nonannouncement days the subset of days that do not feature any 8.30 a.m. are known to market participants well in advance. ET macroeconomic news releases. 9 Since writing this article, I have become aware of a very recent and 11 Strictly speaking, since the ISM index and the minutes are released at somewhat related work by Jubinski and Tomljanovich (forthcoming) that different times, given the intraday volatility pattern displayed by asset prices looks at the intraday response of individual equity prices to FOMC minutes (as documented, for instance, in Andersen and Bollerslev [1997, 1998]), it is for a short, precrisis sample period (2006-07) using a GARCH model. not possible to compare their financial market effects.
FRBNY Economic Policy Review / December 2013 71 C The Volatility of Asset Prices around FOMC Statement, Nonfarm Payrolls, and ISM Manufacturing Index Releases
Panel A: FOMC Statement 1 percent signi cance 5 percent signi cance Standard deviation Standard deviation 0.06 Two-year Treasury yield Ten-year Treasury yield
0.05 0.04 FOMC statement FOMC statement release release 0.04 0.03
0.03 0.02 0.02
0.01 0.01
0 0 0.8 0.3 S&P 500 Euro/U.S. dollar
FOMC statement 0.6 release FOMC statement 0.2 release 0.4
0.1 0.2
0 0 0.30 0.35 Swiss franc/U.S. dollar Japanese yen/U.S. dollar
0.25 0.30 FOMC statement FOMC statement release 0.25 release 0.20 0.20 0.15 0.15 0.10 0.10
0.05 0.05
0 0 13:00 14:00 15:00 13:00 14:00 15:00 Time Time
Source: Author’s calculations. Notes: e chart plots the standard deviation of the ve-minute asset price returns around the news release (solid line) and on control days (the same weekdays and hours of the previous and following weeks of the event day; dashed line). e sample period is January 2005 to March 2011. e interval spans from one hour before to two hours a er the event time. e vertical line signi es the release time of the FOMC statement (see Rosa [2012] for the exact time stamps of the FOMC meetings) in panel A, nonfarm payrolls (8:30 a.m. ET) in panel B, and ISM manufacturing index (10:00 a.m. ET) in panel C. Levene (1960) statistics are employed to test the null hypothesis of equal variances in each subgroup. Dark and white squares denote signi cance of the di erences at the two-sided 1 and 5 percent level, respectively.
72 The Financial Market Effect of FOMC Minutes C () The Volatility of Asset Prices around FOMC Statement, Nonfarm Payrolls, and ISM Manufacturing Index Releases
Panel B: Nonfarm Payrolls 1 percent signi cance 5 percent signi cance Standard deviation Standard deviation 0.08 0.06 Two-year Treasury yieldTen-year Treasury yield
Nonfarm payrolls 0.06 release Nonfarm payrolls 0.04 release 0.04
0.02 0.02
0 0 0.5 0.3 S&P 500 Euro/U.S. dollar
0.4 Nonfarm payrolls release Nonfarm payrolls release 0.2 0.3
0.2 0.1
0.1
0 0
0.4 Swiss franc/ Japanese yen/ U.S. dollar U.S. dollar 0.4 0.3 Nonfarm payrolls Nonfarm payrolls release release 0.3 0.2 0.2
0.1 0.1
0 0.0 7:15 8:00 9:00 10:00 11:00 7:15 8:00 9:00 10:00 11:00 Time Time
FRBNY Economic Policy Review / December 2013 73 C () The Volatility of Asset Prices around FOMC Statement, Nonfarm Payrolls, and ISM Manufacturing Index Releases
Panel C: ISM Manufacturing Index 1 percent signi cance 5 percent signi cance Standard deviation Standard deviation 0.025 0.020 Two-year Ten-year Treasury yield Treasury yield 0.020 0.015 ISM release ISM release 0.015 0.010 0.010
0.005 0.005
0 0
0.30 0.12 S&P 500 Euro/U.S. dollar
0.25 0.10 ISM release ISM release 0.20 0.08 0.15 0.06 0.10
0.04 0.05
0 0.02
0.20 0.20 Swiss franc/U.S. dollar Japanese yen/U.S. dollar
0.15 0.15 ISM release ISM release
0.10 0.10
0.05 0.05
0 0 9:00 10:00 11:00 12:00 9:00 10:00 11:00 12:00 Time Time
74 The Financial Market Effect of FOMC Minutes C The Volatility of Asset Prices around FOMC Minutes Releases: Subsamples
Panel A: January 2005 to December 2007 Subsample 1 percent signi cance 5 percent signi cance Standard deviation Standard deviation 0.025 0.016 Two-year Treasury yield Ten-year Treasury yield 0.020 FOMC minutes 0.012 release FOMC minutes 0.015 release 0.008 0.010
0.004 0.005
0 0 0.20 0.12 S&P 500 Euro/U.S. dollar
0.10 FOMC minutes 0.15 release FOMC minutes 0.08 release
0.10 0.06
0.04 0.05 0.02
0 0 0.12 0.10 Swiss franc/U.S. dollar Japanese yen/U.S. dollar
0.10 FOMC minutes 0.08 release FOMC minutes 0.08 release 0.06 0.06 0.04 0.04
0.02 0.02
0 0 12:55 14:00 15:00 12:55 14:00 15:00 Time Time
Source: Author’s calculations. Notes: e chart plots the standard deviation of ve-minute asset price returns around the FOMC minutes release (solid line) and on control days (the same weekdays and hours of the previous and following weeks of the FOMC minutes release day; dashed line) for two subsamples: January 2005 to December 2007 in panel A and January 2008 to March 2011 in panel B. e interval spans from one hour before to two hours a er the event time. e vertical line signi es the release time of the FOMC minutes, 2 p.m. ET. Levene (1960) statistics are employed to test the null hypothesis of equal variances in each subgroup. Dark and white squares denote signi cance of the di erences at the two-sided 1 and 5 percent level, respectively.
FRBNY Economic Policy Review / December 2013 75 C () The Volatility of Asset Prices around FOMC Minutes Releases: Subsamples
Panel B: January 2008 to March 2011 Subsample 1 percent signi cance 5 percent signi cance Standard deviation Standard deviation 0.015 0.020 Two-year Treasury yield Ten-year Treasury yield
FOMC minutes release 0.015 0.010 FOMC minutes release
0.010
0.005 0.005
0 0
0.35 0.10 S&P 500 Euro/U.S. dollar 0.30 FOMC minutes release 0.08 0.25 FOMC minutes release 0.20 0.06 0.15
0.10 0.04 0.05
0 0.02
0.10 0.10 Swiss franc/U.S. dollar Japanese yen/ U.S. dollar FOMC minutes FOMC minutes 0.08 release 0.08 release
0.06 0.06
0.04 0.04
0.02 0.02 12:55 14:00 15:00 12:55 14:00 15:00 Time Time
76 The Financial Market Effect of FOMC Minutes C Trading Volumes around FOMC Minutes and Statement Releases
1 percent signi cance 5 percent signi cance Panel A: FOMC Minutes Ratio Ratio 4 S&P 500 4 Euro/U.S. dollar
3 3 FOMC minutes FOMC minutes release release 2 2
1 1
0 0
4 Swiss franc/U.S. dollar 4 Japanese yen/U.S. dollar
3 3 FOMC minutes FOMC minutes release release 2 2
1 1
0 0 13:00 14:00 15:00 13:00 14:00 15:00 Time Time Source: Author’s calculations. Notes: e chart plots the median ratio between volumes around the FOMC minutes and statement releases and volumes on control days (the same weekdays and hours of the previous and following weeks of the event days). e sample period is January 2005 to March 2011. e interval spans from one hour before to two hours a er the event time. e vertical line signi es the release time of the FOMC minutes, 2 p.m. ET, in panel A and of FOMC statements (see Rosa [2012] for the exact time stamps of the FOMC meetings) in panel B. e Wilcoxon signed ranks test (Newbold 1988) is employed to test the null hypothesis that the median ratio between ve-minute volumes in the two subgroups equals 1. Dark and white squares denote signi cance of the di erences at the two-sided 1 and 5 percent level, respectively.
4. Robustness Checks minutes) and hours, but on nonannouncement days. Then I test the null hypothesis that the median ratio equals 1, that is, To test the robustness of the baseline results of the previous that the trading activity is the same on days of FOMC minutes section, I also carry out the analysis using trading volumes for releases and nonevent days. The Wilcoxon signed ranks test the S&P 500 stock index and U.S. dollar exchange rates, I redo (see Newbold [1988]) is employed to account for the possibil- the computations on a different subsample, and I look at the ity that the ratio is not normally distributed. Since the existing financial market effect of the release of the Bank of England literature documents a positive contemporaneous relation minutes. between volume and volatility (see Karpoff [1987] and more First, paralleling my earlier analysis on realized volatility, recently Giot, Laurent, and Petitjean [2010] for detailed I examine the relationship between trading volumes and the surveys), I expect that volumes will also respond to the release 12 arrival of news. To adjust for trend growth in trading volumes, of the FOMC minutes and statements. Panel A of Chart 4 and to avoid overweighting the most recent years, for each shows that trading activity is lower than normal before the release day of the FOMC minutes I compute the ratio between 1) the five-minute volumes on release days and the average 12 of 2) the five-minute volumes on the same weekdays (of the For brevity, and because the results are very similar to those on volatility, I provide in a separate appendix (available on request) charts on trading previous and following week of the release day of the FOMC activity around the release of nonfarm payrolls compared with nonevent days.
FRBNY Economic Policy Review / December 2013 77 C () Trading Volumes around FOMC Minutes and Statement Releases
1 percent signi cance 5 percent signi cance Panel B: FOMC Statement
Ratio Ratio 10 10 S&P 500 Euro/U.S. dollar
8 8 FOMC statement FOMC statement release release 6 6
4 4
2 2
0 0 10 Swiss franc/U.S. dollar 10 Japanese yen/U.S. dollar
8 8 FOMC statement FOMC statement release release 6 6
4 4
2 2
0 0 13:00 14:00 15:00 13:00 14:00 15:00 Time Time
release of the minutes, jumps at the time of the release, and regarding its outcome.14 In summary, trading volumes respond then gradually returns to its normal level. The response is most similarly to volatility, with both stock prices and U.S. dollar pronounced for the euro (four times as large as on nonevent exchange rates strongly affected by this monetary news. days) and least pronounced for the S&P 500 (twice as large), Second, I examine whether the asset price response with the Swiss franc and Japanese yen lying in the middle depends on the length of the FOMC meeting, namely, (three times). Panel B indicates that trading activity around whether it is a one- or two-day event. Two-day meetings the release of the FOMC statement strongly and significantly usually provide more time to discuss special topics. I find that increases for all assets, especially for the euro and yen (around the significant increase in volatility is not related to the type of ten times, compared with nonevent days).13 The effect on FOMC meeting (results available upon request). volumes is persistent, and lasts at least one hour and thirty Finally, to show whether the results of increased volatility minutes after the event. Of note, the volume on the S&P 500 is and volume on release days of FOMC minutes hold for especially low before the release time, suggesting highly signif- other central banks, I examine the financial market effect of icant intraday preannouncement effects in the stock market. the Bank of England official communication. I look at two In other words, stock traders restrain from transacting before types of communication: the minutes of the Monetary Policy the news release, and wait for resolution of the uncertainty Committee (MPC) meetings and the Inflation Report. Since November 1998, the minutes have been published at 9:30 a.m.
13 This finding is in line with the intraday results of Fleming and Remolona (1999) for the Treasury market and Fleming and Krishnan (2012) for the 14 This result is consistent with the “calm-before-the-storm” effect Treasury inflation-protected securities market. Fischer and Ranaldo (2011) documented in Jones, Lamont, and Lumsdaine (1998) for macroeconomic show that daily global currency volumes increase on FOMC meeting days. news and Bomfim (2003) for FOMC target rate decisions.
78 The Financial Market Effect of FOMC Minutes London time on the Wednesday of the second week after the 5. Conclusion meetings take place; before that date, the Bank of England minutes were released after the following MPC meeting. The The high-frequency reaction of asset prices to news minutes provide information on the MPC’s assessment of the announcements represents a simple and precise tool for economic outlook and risks, as well as on each Committee assessing how information is impounded into security prices. member’s voting record and assessment of the future This article examines whether, and to what extent, the FOMC monetary policy stance. The Inflation Report is a quarterly minutes contain market-relevant information by looking at publication containing the MPC’s projections for output asset price volatility and trading volume in a narrow window growth and inflation, presented in so-called “fan charts,” around the release of the minutes. The results show that the as well as a detailed analysis of the economic outlook and release significantly affects both the volatility of U.S. asset risks. The report is released at 10:30 a.m. London time and prices and their trading volume. The magnitude of these is accompanied by a press conference. Overall, 147 minutes effects is similar to the financial market effect of a macro- and 49 Inflation Reports have been published between economic release such as the ISM manufacturing index, but January 1999 and March 2011. Consistent with the existing is smaller than the market effect induced by the release of literature, I expect that U.K. assets react to Bank of England the FOMC statement and nonfarm payrolls. The asset price 15 official communication. Estimation results (available upon response to the minutes, however, has declined since 2008, request) show that the volatilities of both five- and ten-year suggesting the greater transparency of the FOMC. U.K. gilts are roughly twice as large at the release time of the minutes compared with nonevent days, and they remain larger than normal for around one hour after the event. Also, the British pound exchange rates (against the euro, U.S. dollar, and Japanese yen) significantly respond to the release of the minutes, whereas the volatility of the FTSE 100 is more muted and not significantly different from volatility during “normal” times. The volatility pattern around the release of the Inflation Report is similar to the pattern displayed around the release of the minutes. The major difference is that the Inflation Report has a stronger immediate effect on asset prices compared with the effect of the minutes. For instance, the volatility of the British pound exchange rate is roughly four times volatility on nonevent days. In summary, the empirical evidence supports the hypothesis that the Bank of England minutes and Inflation Report convey valuable information to investors, with the strongest effect on interest rates and exchange rates.
15 Gerlach-Kristen (2004) documents that the MPC’s voting record contained in the minutes helps predict the future Bank of England policy rate changes. Reeves and Sawicki (2007) investigate the effect of Bank of England communication between 1997 and 2004, and show that the minutes and the Inflation Report significantly affect daily U.K. short-term interest rates.
FRBNY Economic Policy Review / December 2013 79 References
Andersen, T. G., and T. Bollerslev. 1997. “Intraday Periodicity and Danielsson, J., and R. Payne. 2002. “Real Trading Patterns and Prices in Volatility Persistence in Financial Markets.” Journal of Empirical the Spot Foreign Exchange Markets.” Journal of International Finance 4, no. 2-3 (June): 115-58. Money and Finance 21, no. 2 (April): 203-22.
———. 1998. “Deutsche Mark-Dollar Volatility: Intraday Activity Pat- Faust, J., J. H. Rogers, S.-Y. B. Wang, and J. H. Wright. 2007. “The terns, Macroeconomic Announcements, and Longer Run Dependen- High-Frequency Response of Exchange Rates and Interest Rates cies.” Journal of Finance 53, no. 1 (February): 219-65. to Macroeconomic Announcements.” Journal of Monetary Economics 54, no. 4 (May): 1051-68. Apel, M., and M. Blix Grimaldi. 2012. “The Information Content of Central Bank Minutes.” Sveriges Riksbank Working Paper Fischer, A. M., and A. Ranaldo. 2011. “Does FOMC News Increase Series, no. 261, April. Global FX Trading?” Journal of Banking and Finance 35, no. 11 (November): 2965-73. Balduzzi, P., E. J. Elton, and T. C. Green. 2001. “Economic News and Bond Prices: Evidence from the U.S. Treasury Market.” Journal of Fleming, M. J., and N. Krishnan. 2012. “The Microstructure of the Financial and Quantitative Analysis 36, no. 4 (December): TIPS Market.” Federal Reserve Bank of New York Economic 523-43. Policy Review 18, no. 1 (March): 27-45.
Fleming, M. J., and M. Piazzesi. 2005. “Monetary Policy Tick-by-Tick.” Bandi, F., and J. R. Russell. 2008. “Microstructure Noise, Realized Unpublished paper, Federal Reserve Bank of New York, August. Variance, and Optimal Sampling.” Review of Economic Studies 75, no. 2 (April): 339-69. Fleming, M. J., and E. M. Remolona. 1999. “Price Formation and Liquidity in the U.S. Treasury Market: The Response to Public Bernanke, B. S., and K. Kuttner. 2005. “What Explains the Stock Mar- Information.” Journal of Finance 54, no. 5 (October): 1901-15. ket’s Reaction to Federal Reserve Policy?” Journal of Finance 60, no. 3 (June): 1221-57. Gerlach-Kristen, P. 2004. “Is the MPC’s Voting Record Informative about Future U.K. Monetary Policy?” Scandinavian Journal of Bomfim, A. .N 2003. “Pre-Announcement Effects, News Effects, and Economics 106, no. 2: 299-313. Volatility: Monetary Policy and the Stock Market.” Journal of Banking and Finance 27, no. 1 (June): 133-51. Giot, P., S. Laurent, and M. Petitjean. 2010. “Trading Activity, Realized Volatility, and Jumps.” Journal of Empirical Finance 17, no. 1 Boukus, E., and J. V. Rosenberg. 2006. “The Information Content of (January): 168-75. FOMC Minutes.” Unpublished paper, Federal Reserve Bank of New York. Gurkaynak, R., B. Sack, and E. T. Swanson. 2005. “Do Actions Speak Louder than Words? The Response of Asset Prices to Monetary Chaboud, A. P., S. V. Chernenko, E. Howorka, R. S. Krishnasami Iyer, Policy Actions and Statements.” International Journal of D. Liu, and J. H. Wright. 2004. “The High-Frequency Effects of Central Banking 1, no. 1: 55-93. U.S. Macroeconomic Data Releases on Prices and Trading Activ- ity in the Global Interdealer Foreign Exchange Market.” Board Jones, C., O. Lamont, and R. Lumsdaine. 1998. “Macroeconomic News of Governors of the Federal Reserve System International and Bond Market Volatility.” Journal of Financial Economics Finance Discussion Papers, no. 823, November. 47, no. 3 (March): 315-37.
Chaboud, A. P., S. V. Chernenko, and J. H. Wright. 2008. “Trading Activity Jubinski, D., and M. Tomljanovich. Forthcoming. “Do FOMC Min- and Macroeconomic Announcements in High-Frequency Exchange utes Matter to Markets? An Intraday Analysis of FOMC Minutes Rate Data.” Journal of the European Economic Association 6, Releases on Individual Equity Volatility and Returns.” Review of no. 2-3 (April-May): 589-96. Financial Economics.
80 The Financial Market Effect of FOMC Minutes References (Continued)
Karpoff, J. .M 1987. “The Relation between Price Changes and Trading Rigobon, R., and B. P. Sack. 2004. “The Impact of Monetary Policy Volume: A Survey.” Journal of Financial and Quantitative on Asset Prices.” Journal of Monetary Economics 51, no. 8 Analysis 22, no. 1 (March): 109-26. (November): 1553-75.
Kohn, D. L., and B. Sack. 2004. “Central Bank Talk: Does It Matter and Rosa, C. 2011a. “Words that Shake Traders: The Stock Market’s Reac- Why?” In Macroeconomics, Monetary Policy, and Financial tion to Central Bank Communication in Real Time.” Journal of Stability, 175-206. Proceedings of a Conference Held at the Bank Empirical Finance 18, no. 5 (December): 915-34. of Canada, Ottawa, Canada, June 2003. ———. 2011b. “The High-Frequency Response of Exchange Rates to Kuttner, K. 2001. “Monetary Policy Surprises and Interest Rates: Monetary Policy Actions and Statements.” Journal of Banking Evidence from the Fed Funds Futures Markets.” Journal of and Finance 35, no. 2 (February): 478-89. Monetary Economics 47, no. 3 (June): 523-44. ———. 2012. “How ‘Unconventional’ Are Large-Scale Asset Purchases? Levene, H. 1960. “Robust Tests for Equality of Variances.” In I. Olkin, The Impact of Monetary Policy on Asset Prices.” Federal Reserve ed., Contributions to Probability and Statistics. Palo Alto, Bank of New York Staff Reports, no 560, May. CA: Stanford University Press. Swanson, E. T., and J. C. Williams. 2012. “Measuring the Effect of Newbold, P. 1988. Statistics for Business and Economics. 2d. ed. the Zero Lower Bound on Medium- and Longer-Term Interest New York: Prentice Hall. Rates.” Federal Reserve Bank of San Francisco Working Paper no. 2012-02, May. Phylaktis, K., and L. Chen. 2009. “Price Discovery in Foreign Exchange Markets: A Comparison of Indicative and Actual Transaction Prices.” Woodford, M. 2005. “Central Bank Communication and Policy Journal of Empirical Finance 16, no. 4 (September): 640-54. Effectiveness.” NBER Working Paper no. 11898, December.
Reeves, R., and M. Sawicki. 2007. “Do Financial Markets React to Bank of England Communication?” European Journal of Political Economy 23, no. 1 (March): 207-27.
The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.
FRBNY Economic Policy Review / December 2013 81 Federal Reserve Bank of New York Economic Policy Review
Economic Policy Review articles may be reproduced for educational or training purposes, provided they are reprinted in full; include credit to the author(s), the publication, and the Bank; and include the publication’s disclaimer.
Library of Congress Card Number: 77-646559
Federal Reserve Bank of New York research publications—the Economic Policy Review, Current Issues in Economics and Finance, Staff Reports, Research Update, and the annual publications catalogue—are available at no charge from the Research Group’s website. www.newyorkfed.org/research Follow us on Twitter: @NYFedResearch